Added 0.17.2 release notes (#6028)

### What problem does this PR solve?



### Type of change

- [x] Documentation Update

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -5,11 +5,17 @@ labels: [bug]
 body:
 - type: checkboxes
   attributes:
-    label: Is there an existing issue for the same bug?
-    description: Please check if an issue already exists for the bug you encountered.
+    label: Self Checks
+    description: "Please check the following in order to be responded in time :)"
     options:
-    - label: I have checked the existing issues.
-      required: true
+      - label: I have searched for existing issues [search for existing issues](https://github.com/infiniflow/ragflow/issues), including closed ones.
+        required: true
+      - label: I confirm that I am using English to submit this report ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+        required: true
+      - label: Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+        required: true
+      - label: "Please do not modify this template :) and fill in all the required fields."
+        required: true
 - type: markdown
   attributes:
     value: "Please provide the following information to help us understand the issue."

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -5,10 +5,16 @@ labels: [feature request]
 body:
   - type: checkboxes
     attributes:
-      label: Is there an existing issue for the same feature request?
-      description: Please check if an issue already exists for the feature you request.
+      label: Self Checks
+      description: "Please check the following in order to be responded in time :)"
       options:
-        - label: I have checked the existing issues.
+        - label: I have searched for existing issues [search for existing issues](https://github.com/infiniflow/ragflow/issues), including closed ones.
+          required: true
+        - label: I confirm that I am using English to submit this report ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+          required: true
+        - label: Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+          required: true
+        - label: "Please do not modify this template :) and fill in all the required fields."
           required: true
   - type: textarea
     attributes:

.github/ISSUE_TEMPLATE/question.yml

@@ -3,6 +3,19 @@ description: Ask questions on RAGFlow
 title: "[Question]: "
 labels: [question]
 body:
+- type: checkboxes
+  attributes:
+    label: Self Checks
+    description: "Please check the following in order to be responded in time :)"
+    options:
+      - label: I have searched for existing issues [search for existing issues](https://github.com/infiniflow/ragflow/issues), including closed ones.
+        required: true
+      - label: I confirm that I am using English to submit this report ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+        required: true
+      - label: Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+        required: true
+      - label: "Please do not modify this template :) and fill in all the required fields."
+        required: true
 - type: markdown
   attributes:
     value: |

.github/workflows/release.yml

@@ -75,12 +75,6 @@ jobs:
           # The body field does not support environment variable substitution directly.
           body_path: release_body.md
 
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
-
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-
       # https://github.com/marketplace/actions/docker-login
       - name: Login to Docker Hub
         uses: docker/login-action@v3

.github/workflows/tests.yml

@@ -32,12 +32,9 @@ jobs:
       # https://github.com/hmarr/debug-action
       #- uses: hmarr/debug-action@v2
 
-      - name: Show PR labels
+      - name: Show who triggered this workflow
         run: |
           echo "Workflow triggered by ${{ github.event_name }}"
-          if [[ ${{ github.event_name }} == 'pull_request' ]]; then
-            echo "PR labels: ${{ join(github.event.pull_request.labels.*.name, ', ') }}"
-          fi
 
       - name: Ensure workspace ownership
         run: echo "chown -R $USER $GITHUB_WORKSPACE" && sudo chown -R $USER $GITHUB_WORKSPACE
@@ -68,7 +65,7 @@ jobs:
 
       - name: Start ragflow:nightly-slim
         run: |
-          echo "RAGFLOW_IMAGE=infiniflow/ragflow:nightly-slim" >> docker/.env
+          echo -e "\nRAGFLOW_IMAGE=infiniflow/ragflow:nightly-slim" >> docker/.env
           sudo docker compose -f docker/docker-compose.yml up -d
 
       - name: Stop ragflow:nightly-slim
@@ -78,7 +75,7 @@ jobs:
 
       - name: Start ragflow:nightly
         run: |
-          echo "RAGFLOW_IMAGE=infiniflow/ragflow:nightly" >> docker/.env
+          echo -e "\nRAGFLOW_IMAGE=infiniflow/ragflow:nightly" >> docker/.env
           sudo docker compose -f docker/docker-compose.yml up -d
 
       - name: Run sdk tests against Elasticsearch
@@ -99,8 +96,17 @@ jobs:
             echo "Waiting for service to be available..."
             sleep 5
           done
-          cd sdk/python && uv sync --python 3.10 --frozen && uv pip install . && source .venv/bin/activate && cd test/test_frontend_api && pytest -s --tb=short get_email.py test_dataset.py 
+          cd sdk/python && uv sync --python 3.10 --frozen && uv pip install . && source .venv/bin/activate && cd test/test_frontend_api && pytest -s --tb=short get_email.py test_dataset.py
           
+      - name: Run http api tests against Elasticsearch
+        run: |
+          export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
+          export HOST_ADDRESS=http://host.docker.internal:9380
+          until sudo docker exec ragflow-server curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+            echo "Waiting for service to be available..."
+            sleep 5
+          done
+          cd sdk/python && uv sync --python 3.10 --frozen && uv pip install . && source .venv/bin/activate && cd test/test_http_api && pytest -s --tb=short -m "not slow"
 
       - name: Stop ragflow:nightly
         if: always()  # always run this step even if previous steps failed
@@ -131,6 +137,16 @@ jobs:
           done
           cd sdk/python && uv sync --python 3.10 --frozen && uv pip install . && source .venv/bin/activate && cd test/test_frontend_api && pytest -s --tb=short get_email.py test_dataset.py
 
+      - name: Run http api tests against Infinity
+        run: |
+          export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
+          export HOST_ADDRESS=http://host.docker.internal:9380
+          until sudo docker exec ragflow-server curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+            echo "Waiting for service to be available..."
+            sleep 5
+          done
+          cd sdk/python && uv sync --python 3.10 --frozen && uv pip install . && source .venv/bin/activate && cd test/test_http_api && pytest -s --tb=short -m "not slow"
+
       - name: Stop ragflow:nightly
         if: always()  # always run this step even if previous steps failed
         run: |

.gitignore

@@ -38,3 +38,7 @@ sdk/python/dist/
 sdk/python/ragflow_sdk.egg-info/
 huggingface.co/
 nltk_data/
+
+# Exclude hash-like temporary files like 9b5ad71b2ce5302211f9c61530b329a4922fc6a4
+*[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]*
+.lh/

Dockerfile

@@ -59,14 +59,15 @@ RUN --mount=type=cache,id=ragflow_apt,target=/var/cache/apt,sharing=locked \
     apt install -y default-jdk && \
     apt install -y libatk-bridge2.0-0 && \
     apt install -y libpython3-dev libgtk-4-1 libnss3 xdg-utils libgbm-dev && \
+    apt install -y libjemalloc-dev && \
     apt install -y python3-pip pipx nginx unzip curl wget git vim less
 
 RUN if [ "$NEED_MIRROR" == "1" ]; then \
-        pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple && \
-        pip3 config set global.trusted-host pypi.tuna.tsinghua.edu.cn; \
+        pip3 config set global.index-url https://mirrors.aliyun.com/pypi/simple && \
+        pip3 config set global.trusted-host mirrors.aliyun.com; \
         mkdir -p /etc/uv && \
         echo "[[index]]" > /etc/uv/uv.toml && \
-        echo 'url = "https://pypi.tuna.tsinghua.edu.cn/simple"' >> /etc/uv/uv.toml && \
+        echo 'url = "https://mirrors.aliyun.com/pypi/simple"' >> /etc/uv/uv.toml && \
         echo "default = true" >> /etc/uv/uv.toml; \
     fi; \
     pipx install uv
@@ -150,9 +151,9 @@ COPY pyproject.toml uv.lock ./
 # uv records index url into uv.lock but doesn't failover among multiple indexes
 RUN --mount=type=cache,id=ragflow_uv,target=/root/.cache/uv,sharing=locked \
     if [ "$NEED_MIRROR" == "1" ]; then \
-        sed -i 's|pypi.org|pypi.tuna.tsinghua.edu.cn|g' uv.lock; \
+        sed -i 's|pypi.org|mirrors.aliyun.com/pypi|g' uv.lock; \
     else \
-        sed -i 's|pypi.tuna.tsinghua.edu.cn|pypi.org|g' uv.lock; \
+        sed -i 's|mirrors.aliyun.com/pypi|pypi.org|g' uv.lock; \
     fi; \
     if [ "$LIGHTEN" == "1" ]; then \
         uv sync --python 3.10 --frozen; \
@@ -196,6 +197,7 @@ COPY deepdoc deepdoc
 COPY rag rag
 COPY agent agent
 COPY graphrag graphrag
+COPY agentic_reasoning agentic_reasoning
 COPY pyproject.toml uv.lock ./
 
 COPY docker/service_conf.yaml.template ./conf/service_conf.yaml.template

README.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -78,9 +78,10 @@ Try our demo at [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Latest Updates
 
+- 2025-02-28 Combined with Internet search (Tavily), supports reasoning like Deep Research for any LLMs.
 - 2025-02-05 Updates the model list of 'SILICONFLOW' and adds support for Deepseek-R1/DeepSeek-V3.
 - 2025-01-26 Optimizes knowledge graph extraction and application, offering various configuration options.
-- 2024-12-18 Upgrades Document Layout Analysis model in Deepdoc.
+- 2024-12-18 Upgrades Document Layout Analysis model in DeepDoc.
 - 2024-12-04 Adds support for pagerank score in knowledge base.
 - 2024-11-22 Adds more variables to Agent.
 - 2024-11-01 Adds keyword extraction and related question generation to the parsed chunks to improve the accuracy of retrieval.
@@ -173,19 +174,23 @@ releases! 🌟
 
 3. Start up the server using the pre-built Docker images:
 
-   > The command below downloads the `v0.16.0-slim` edition of the RAGFlow Docker image. Refer to the following table for descriptions of different RAGFlow editions. To download an RAGFlow edition different from `v0.16.0-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0` for the full edition `v0.16.0`.
+> [!CAUTION]
+> All Docker images are built for x86 platforms. We don't currently offer Docker images for ARM64.
+> If you are on an ARM64 platform, follow [this guide](https://ragflow.io/docs/dev/build_docker_image) to build a Docker image compatible with your system.
+
+   > The command below downloads the `v0.17.2-slim` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.17.2-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2` for the full edition `v0.17.2`.
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    |-------------------|-----------------|-----------------------|--------------------------|
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                     | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                   | Stable release            |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
-   | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
+   | nightly-slim      | &approx;2       | ❌                   | _Unstable_ nightly build  |
 
 4. Check the server status after having the server up and running:
 
@@ -204,9 +209,6 @@ releases! 🌟
      /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
    > If you skip this confirmation step and directly log in to RAGFlow, your browser may prompt a `network anormal`
@@ -240,7 +242,7 @@ to `<YOUR_SERVING_PORT>:80`.
 Updates to the above configurations require a reboot of all containers to take effect:
 
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ### Switch doc engine from Elasticsearch to Infinity
@@ -253,12 +255,15 @@ RAGFlow uses Elasticsearch by default for storing full text and vectors. To swit
    $ docker compose -f docker/docker-compose.yml down -v
    ```
 
+> [!WARNING]
+> `-v` will delete the docker container volumes, and the existing data will be cleared.
+
 2. Set `DOC_ENGINE` in **docker/.env** to `infinity`.
 
 3. Start the containers:
 
    ```bash
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ docker compose -f docker-compose.yml up -d
    ```
 
 > [!WARNING]

README_id.md

@@ -22,7 +22,7 @@
         <img alt="Lencana Daring" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Rilis%20Terbaru" alt="Rilis Terbaru">
@@ -41,7 +41,7 @@
 </h4>
 
 <details open>
-<summary></b>📕 Daftar Isi</b></summary>
+<summary><b>📕 Daftar Isi </b> </summary>
 
 - 💡 [Apa Itu RAGFlow?](#-apa-itu-ragflow)
 - 🎮 [Demo](#-demo)
@@ -75,9 +75,10 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Pembaruan Terbaru
 
+- 2025-02-28 dikombinasikan dengan pencarian Internet (TAVILY), mendukung penelitian mendalam untuk LLM apa pun.
 - 2025-02-05 Memperbarui daftar model 'SILICONFLOW' dan menambahkan dukungan untuk Deepseek-R1/DeepSeek-V3.
 - 2025-01-26 Optimalkan ekstraksi dan penerapan grafik pengetahuan dan sediakan berbagai opsi konfigurasi.
-- 2024-12-18 Meningkatkan model Analisis Tata Letak Dokumen di Deepdoc.
+- 2024-12-18 Meningkatkan model Analisis Tata Letak Dokumen di DeepDoc.
 - 2024-12-04 Mendukung skor pagerank ke basis pengetahuan.
 - 2024-11-22 Peningkatan definisi dan penggunaan variabel di Agen.
 - 2024-11-01 Penambahan ekstraksi kata kunci dan pembuatan pertanyaan terkait untuk meningkatkan akurasi pengambilan.
@@ -166,21 +167,25 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
 
 3. Bangun image Docker pre-built dan jalankan server:
 
-   > Perintah di bawah ini mengunduh edisi v0.16.0-slim dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.16.0-slim, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server. Misalnya, atur RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0 untuk edisi lengkap v0.16.0.
+> [!CAUTION]
+> Semua gambar Docker dibangun untuk platform x86. Saat ini, kami tidak menawarkan gambar Docker untuk ARM64.
+> Jika Anda menggunakan platform ARM64, [silakan gunakan panduan ini untuk membangun gambar Docker yang kompatibel dengan sistem Anda](https://ragflow.io/docs/dev/build_docker_image).
+
+   > Perintah di bawah ini mengunduh edisi v0.17.2-slim dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.17.2-slim, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server. Misalnya, atur RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2 untuk edisi lengkap v0.17.2.
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
 
-4. Periksa status server setelah server aktif dan berjalan:
+1. Periksa status server setelah server aktif dan berjalan:
 
    ```bash
    $ docker logs -f ragflow-server
@@ -197,18 +202,15 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
      /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
    > Jika Anda melewatkan langkah ini dan langsung login ke RAGFlow, browser Anda mungkin menampilkan error `network anormal`
    > karena RAGFlow mungkin belum sepenuhnya siap.
 
-5. Buka browser web Anda, masukkan alamat IP server Anda, dan login ke RAGFlow.
+2. Buka browser web Anda, masukkan alamat IP server Anda, dan login ke RAGFlow.
    > Dengan pengaturan default, Anda hanya perlu memasukkan `http://IP_DEVICE_ANDA` (**tanpa** nomor port) karena
    > port HTTP default `80` bisa dihilangkan saat menggunakan konfigurasi default.
-6. Dalam [service_conf.yaml.template](./docker/service_conf.yaml.template), pilih LLM factory yang diinginkan di `user_default_llm` dan perbarui
+3. Dalam [service_conf.yaml.template](./docker/service_conf.yaml.template), pilih LLM factory yang diinginkan di `user_default_llm` dan perbarui
    bidang `API_KEY` dengan kunci API yang sesuai.
 
    > Lihat [llm_api_key_setup](https://ragflow.io/docs/dev/llm_api_key_setup) untuk informasi lebih lanjut.
@@ -230,7 +232,7 @@ menjadi `<YOUR_SERVING_PORT>:80`.
 Pembaruan konfigurasi ini memerlukan reboot semua kontainer agar efektif:
 
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ## 🔧 Membangun Docker Image tanpa Model Embedding

README_ja.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -55,9 +55,10 @@
 
 ## 🔥 最新情報
 
+- 2025-02-28 インターネット検索 (TAVILY) と組み合わせて、あらゆる LLM の詳細な調査をサポートします。
 - 2025-02-05 シリコン フローの St およびモデル リストを更新し、Deep Seek-R1/Deep Seek-V3 のサポートを追加しました。
 - 2025-01-26 ナレッジ グラフの抽出と適用を最適化し、さまざまな構成オプションを提供します。
-- 2024-12-18 Deepdoc のドキュメント レイアウト分析モデルをアップグレードします。
+- 2024-12-18 DeepDoc のドキュメント レイアウト分析モデルをアップグレードします。
 - 2024-12-04 ナレッジ ベースへのページランク スコアをサポートしました。
 - 2024-11-22 エージェントでの変数の定義と使用法を改善しました。
 - 2024-11-01 再現の精度を向上させるために、解析されたチャンクにキーワード抽出と関連質問の生成を追加しました。
@@ -146,21 +147,25 @@
 
 3. ビルド済みの Docker イメージをビルドし、サーバーを起動する:
 
-   > 以下のコマンドは、RAGFlow Docker イメージの v0.16.0-slim エディションをダウンロードします。異なる RAGFlow エディションの説明については、以下の表を参照してください。v0.16.0-slim とは異なるエディションをダウンロードするには、docker/.env ファイルの RAGFLOW_IMAGE 変数を適宜更新し、docker compose を使用してサーバーを起動してください。例えば、完全版 v0.16.0 をダウンロードするには、RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0 と設定します。
+> [!CAUTION]
+> 現在、公式に提供されているすべての Docker イメージは x86 アーキテクチャ向けにビルドされており、ARM64 用の Docker イメージは提供されていません。
+> ARM64 アーキテクチャのオペレーティングシステムを使用している場合は、[このドキュメント](https://ragflow.io/docs/dev/build_docker_image)を参照して Docker イメージを自分でビルドしてください。
+
+   > 以下のコマンドは、RAGFlow Docker イメージの v0.17.2-slim エディションをダウンロードします。異なる RAGFlow エディションの説明については、以下の表を参照してください。v0.17.2-slim とは異なるエディションをダウンロードするには、docker/.env ファイルの RAGFLOW_IMAGE 変数を適宜更新し、docker compose を使用してサーバーを起動してください。例えば、完全版 v0.17.2 をダウンロードするには、RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2 と設定します。
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
-   | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
+   | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 
-4. サーバーを立ち上げた後、サーバーの状態を確認する:
+1. サーバーを立ち上げた後、サーバーの状態を確認する:
 
    ```bash
    $ docker logs -f ragflow-server
@@ -176,16 +181,13 @@
     /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
    > もし確認ステップをスキップして直接 RAGFlow にログインした場合、その時点で RAGFlow が完全に初期化されていない可能性があるため、ブラウザーがネットワーク異常エラーを表示するかもしれません。
 
-5. ウェブブラウザで、プロンプトに従ってサーバーの IP アドレスを入力し、RAGFlow にログインします。
+2. ウェブブラウザで、プロンプトに従ってサーバーの IP アドレスを入力し、RAGFlow にログインします。
    > デフォルトの設定を使用する場合、デフォルトの HTTP サービングポート `80` は省略できるので、与えられたシナリオでは、`http://IP_OF_YOUR_MACHINE`（ポート番号は省略）だけを入力すればよい。
-6. [service_conf.yaml.template](./docker/service_conf.yaml.template) で、`user_default_llm` で希望の LLM ファクトリを選択し、`API_KEY` フィールドを対応する API キーで更新する。
+3. [service_conf.yaml.template](./docker/service_conf.yaml.template) で、`user_default_llm` で希望の LLM ファクトリを選択し、`API_KEY` フィールドを対応する API キーで更新する。
 
    > 詳しくは [llm_api_key_setup](https://ragflow.io/docs/dev/llm_api_key_setup) を参照してください。
 
@@ -208,7 +210,7 @@
 > すべてのシステム設定のアップデートを有効にするには、システムの再起動が必要です:
 >
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ### Elasticsearch から Infinity にドキュメントエンジンを切り替えます
@@ -219,11 +221,12 @@ RAGFlow はデフォルトで Elasticsearch を使用して全文とベクトル
    ```bash
    $ docker compose -f docker/docker-compose.yml down -v
    ```
+   Note: `-v` は docker コンテナのボリュームを削除し、既存のデータをクリアします。
 2. **docker/.env** の「DOC \_ ENGINE」を「infinity」に設定します。
 
 3. 起動コンテナ：
    ```bash
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ docker compose -f docker-compose.yml up -d
    ```
    > [!WARNING]  
    > Linux/arm64 マシンでの Infinity への切り替えは正式にサポートされていません。

README_ko.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -55,9 +55,10 @@
 
 ## 🔥 업데이트
 
+- 2025-02-28 인터넷 검색(TAVILY)과 결합되어 모든 LLM에 대한 심층 연구를 지원합니다.
 - 2025-02-05 'SILICONFLOW' 모델 목록을 업데이트하고 Deepseek-R1/DeepSeek-V3에 대한 지원을 추가합니다.
 - 2025-01-26 지식 그래프 추출 및 적용을 최적화하고 다양한 구성 옵션을 제공합니다.
-- 2024-12-18 Deepdoc의 문서 레이아웃 분석 모델 업그레이드.
+- 2024-12-18 DeepDoc의 문서 레이아웃 분석 모델 업그레이드.
 - 2024-12-04 지식베이스에 대한 페이지랭크 점수를 지원합니다.
 
 - 2024-11-22 에이전트의 변수 정의 및 사용을 개선했습니다.
@@ -147,21 +148,25 @@
 
 3. 미리 빌드된 Docker 이미지를 생성하고 서버를 시작하세요:
 
-   > 아래 명령어는 RAGFlow Docker 이미지의 v0.16.0-slim 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.16.0-slim과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오. 예를 들어, 전체 버전인 v0.16.0을 다운로드하려면 RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0로 설정합니다.
+> [!CAUTION]
+> 모든 Docker 이미지는 x86 플랫폼을 위해 빌드되었습니다. 우리는 현재 ARM64 플랫폼을 위한 Docker 이미지를 제공하지 않습니다.
+> ARM64 플랫폼을 사용 중이라면, [시스템과 호환되는 Docker 이미지를 빌드하려면 이 가이드를 사용해 주세요](https://ragflow.io/docs/dev/build_docker_image).
+
+   > 아래 명령어는 RAGFlow Docker 이미지의 v0.17.2-slim 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.17.2-slim과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오. 예를 들어, 전체 버전인 v0.17.2을 다운로드하려면 RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2로 설정합니다.
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
-   | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
+   | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 
-4. 서버가 시작된 후 서버 상태를 확인하세요:
+1. 서버가 시작된 후 서버 상태를 확인하세요:
 
    ```bash
    $ docker logs -f ragflow-server
@@ -177,16 +182,13 @@
     /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
    > 만약 확인 단계를 건너뛰고 바로 RAGFlow에 로그인하면, RAGFlow가 완전히 초기화되지 않았기 때문에 브라우저에서 `network anormal` 오류가 발생할 수 있습니다.
 
-5. 웹 브라우저에 서버의 IP 주소를 입력하고 RAGFlow에 로그인하세요.
+2. 웹 브라우저에 서버의 IP 주소를 입력하고 RAGFlow에 로그인하세요.
    > 기본 설정을 사용할 경우, `http://IP_OF_YOUR_MACHINE`만 입력하면 됩니다 (포트 번호는 제외). 기본 HTTP 서비스 포트 `80`은 기본 구성으로 사용할 때 생략할 수 있습니다.
-6. [service_conf.yaml.template](./docker/service_conf.yaml.template) 파일에서 원하는 LLM 팩토리를 `user_default_llm`에 선택하고, `API_KEY` 필드를 해당 API 키로 업데이트하세요.
+3. [service_conf.yaml.template](./docker/service_conf.yaml.template) 파일에서 원하는 LLM 팩토리를 `user_default_llm`에 선택하고, `API_KEY` 필드를 해당 API 키로 업데이트하세요.
 
    > 자세한 내용은 [llm_api_key_setup](https://ragflow.io/docs/dev/llm_api_key_setup)를 참조하세요.
 
@@ -209,7 +211,7 @@
 > 모든 시스템 구성 업데이트는 적용되기 위해 시스템 재부팅이 필요합니다.
 >
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ### Elasticsearch 에서 Infinity 로 문서 엔진 전환
@@ -220,6 +222,7 @@ RAGFlow 는 기본적으로 Elasticsearch 를 사용하여 전체 텍스트 및
    ```bash
    $docker compose-f docker/docker-compose.yml down -v
    ```
+   Note: `-v` 는 docker 컨테이너의 볼륨을 삭제하고 기존 데이터를 지우며, 이 작업은 컨테이너를 중지하는 것과 동일합니다.
 2. **docker/.env**의 "DOC_ENGINE" 을 "infinity" 로 설정합니다.
 3. 컨테이너 부팅:
    ```bash

README_pt_br.md

@@ -22,7 +22,7 @@
         <img alt="Badge Estático" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Última%20Relese" alt="Última Versão">
@@ -41,7 +41,7 @@
 </h4>
 
 <details open>
-<summary></b>📕 Índice</b></summary>
+<summary><b>📕 Índice</b></summary>
 
 - 💡 [O que é o RAGFlow?](#-o-que-é-o-ragflow)
 - 🎮 [Demo](#-demo)
@@ -75,9 +75,10 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Últimas Atualizações
 
+- 28/02/2025 combinado com a pesquisa na Internet (T AVI LY), suporta pesquisas profundas para qualquer LLM.
 - 05-02-2025 Atualiza a lista de modelos de 'SILICONFLOW' e adiciona suporte para Deepseek-R1/DeepSeek-V3.
 - 26-01-2025 Otimize a extração e aplicação de gráficos de conhecimento e forneça uma variedade de opções de configuração.
-- 18-12-2024 Atualiza o modelo de Análise de Layout de Documentos no Deepdoc.
+- 18-12-2024 Atualiza o modelo de Análise de Layout de Documentos no DeepDoc.
 - 04-12-2024 Adiciona suporte para pontuação de pagerank na base de conhecimento.
 - 22-11-2024 Adiciona mais variáveis para o Agente.
 - 01-11-2024 Adiciona extração de palavras-chave e geração de perguntas relacionadas aos blocos analisados para melhorar a precisão da recuperação.
@@ -166,19 +167,23 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 
 3.  Inicie o servidor usando as imagens Docker pré-compiladas:
 
-    > O comando abaixo baixa a edição `v0.16.0-slim` da imagem Docker do RAGFlow. Consulte a tabela a seguir para descrições de diferentes edições do RAGFlow. Para baixar uma edição do RAGFlow diferente da `v0.16.0-slim`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor. Por exemplo: defina `RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0` para a edição completa `v0.16.0`.
+> [!CAUTION]
+> Todas as imagens Docker são construídas para plataformas x86. Atualmente, não oferecemos imagens Docker para ARM64.
+> Se você estiver usando uma plataforma ARM64, por favor, utilize [este guia](https://ragflow.io/docs/dev/build_docker_image) para construir uma imagem Docker compatível com o seu sistema.
+
+    > O comando abaixo baixa a edição `v0.17.2-slim` da imagem Docker do RAGFlow. Consulte a tabela a seguir para descrições de diferentes edições do RAGFlow. Para baixar uma edição do RAGFlow diferente da `v0.17.2-slim`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor. Por exemplo: defina `RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2` para a edição completa `v0.17.2`.
 
     ```bash
-    $ cd ragflow
-    $ docker compose -f docker/docker-compose.yml up -d
+    $ cd ragflow/docker
+    $ docker compose -f docker-compose.yml up -d
     ```
 
     | Tag da imagem RAGFlow | Tamanho da imagem (GB) | Possui modelos de incorporação? | Estável?                 |
     | --------------------- | ---------------------- | ------------------------------- | ------------------------ |
-    | v0.16.0               | ~9                     | :heavy_check_mark:              | Lançamento estável       |
-    | v0.16.0-slim          | ~2                     | ❌                              | Lançamento estável       |
+    | v0.17.2               | ~9                     | :heavy_check_mark:              | Lançamento estável       |
+    | v0.17.2-slim          | ~2                     | ❌                              | Lançamento estável       |
     | nightly               | ~9                     | :heavy_check_mark:              | _Instável_ build noturno |
-    | nightly-slim          | ~2                     | ❌                              | _Instável_ build noturno |
+    | nightly-slim          | ~2                     | ❌                               | _Instável_ build noturno |
 
 4.  Verifique o status do servidor após tê-lo iniciado:
 
@@ -196,9 +201,6 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
      /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
      * Rodando em todos os endereços (0.0.0.0)
-     * Rodando em http://127.0.0.1:9380
-     * Rodando em http://x.x.x.x:9380
-     INFO:werkzeug:Pressione CTRL+C para sair
     ```
 
     > Se você pular essa etapa de confirmação e acessar diretamente o RAGFlow, seu navegador pode exibir um erro `network anormal`, pois, nesse momento, seu RAGFlow pode não estar totalmente inicializado.
@@ -228,7 +230,7 @@ Para atualizar a porta HTTP de serviço padrão (80), vá até [docker-compose.y
 Atualizações nas configurações acima exigem um reinício de todos os contêineres para que tenham efeito:
 
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ### Mudar o mecanismo de documentos de Elasticsearch para Infinity
@@ -240,13 +242,13 @@ O RAGFlow usa o Elasticsearch por padrão para armazenar texto completo e vetore
    ```bash
    $ docker compose -f docker/docker-compose.yml down -v
    ```
-
+   Note: `-v` irá deletar os volumes do contêiner, e os dados existentes serão apagados.
 2. Defina `DOC_ENGINE` no **docker/.env** para `infinity`.
 
 3. Inicie os contêineres:
 
    ```bash
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ docker compose -f docker-compose.yml up -d
    ```
 
 > [!ATENÇÃO]

README_tzh.md

@@ -21,7 +21,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -54,9 +54,10 @@
 
 ## 🔥 近期更新
 
+- 2025-02-28 結合網路搜尋（Tavily），對於任意大模型實現類似 Deep Research 的推理功能.
 - 2025-02-05 更新「SILICONFLOW」的型號清單並新增 Deepseek-R1/DeepSeek-V3 的支援。
 - 2025-01-26 最佳化知識圖譜的擷取與應用，提供了多種配置選擇。
-- 2024-12-18 升級了 Deepdoc 的文檔佈局分析模型。
+- 2024-12-18 升級了 DeepDoc 的文檔佈局分析模型。
 - 2024-12-04 支援知識庫的 Pagerank 分數。
 - 2024-11-22 完善了 Agent 中的變數定義和使用。
 - 2024-11-01 對解析後的 chunk 加入關鍵字抽取和相關問題產生以提高回想的準確度。
@@ -145,19 +146,23 @@
 
 3. 進入 **docker** 資料夾，利用事先編譯好的 Docker 映像啟動伺服器：
 
-   > 執行以下指令會自動下載 RAGFlow slim Docker 映像 `v0.16.0-slim`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.16.0-slim` 的 Docker 映像，請在執行 `docker compose` 啟動服務之前先更新 **docker/.env** 檔案內的 `RAGFLOW_IMAGE` 變數。例如，你可以透過設定 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0` 來下載 RAGFlow 鏡像的 `v0.16.0` 完整發行版。
+> [!CAUTION]
+> 所有 Docker 映像檔都是為 x86 平台建置的。目前，我們不提供 ARM64 平台的 Docker 映像檔。
+> 如果您使用的是 ARM64 平台，請使用 [這份指南](https://ragflow.io/docs/dev/build_docker_image) 來建置適合您系統的 Docker 映像檔。
+
+   > 執行以下指令會自動下載 RAGFlow slim Docker 映像 `v0.17.2-slim`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.17.2-slim` 的 Docker 映像，請在執行 `docker compose` 啟動服務之前先更新 **docker/.env** 檔案內的 `RAGFLOW_IMAGE` 變數。例如，你可以透過設定 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2` 來下載 RAGFlow 鏡像的 `v0.17.2` 完整發行版。
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
-   | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
+   | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 
    > [!TIP]
    > 如果你遇到 Docker 映像檔拉不下來的問題，可以在 **docker/.env** 檔案內根據變數 `RAGFLOW_IMAGE` 的註解提示選擇華為雲或阿里雲的對應映像。
@@ -181,9 +186,6 @@
     /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
    > 如果您跳過這一步驟系統確認步驟就登入 RAGFlow，你的瀏覽器有可能會提示 `network anormal` 或 `網路異常`，因為 RAGFlow 可能並未完全啟動成功。
@@ -200,7 +202,7 @@
 
 系統配置涉及以下三份文件：
 
-- [.env](./docker/.env)：存放一些基本的系統環境變量，例如 `SVR_HTTP_PORT`、`MYSQL_PASSWORD`、`MINIO_PASSWORD` 等。
+- [.env](./docker/.env)：存放一些系統環境變量，例如 `SVR_HTTP_PORT`、`MYSQL_PASSWORD`、`MINIO_PASSWORD` 等。
 - [service_conf.yaml.template](./docker/service_conf.yaml.template)：設定各類別後台服務。
 - [docker-compose.yml](./docker/docker-compose.yml): 系統依賴該檔案完成啟動。
 
@@ -215,7 +217,7 @@
 > 所有系統配置都需要透過系統重新啟動生效：
 >
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ###把文檔引擎從 Elasticsearch 切換成為 Infinity
@@ -227,13 +229,14 @@ RAGFlow 預設使用 Elasticsearch 儲存文字和向量資料. 如果要切換
    ```bash
    $ docker compose -f docker/docker-compose.yml down -v
    ```
+   Note: `-v` 將會刪除 docker 容器的 volumes，已有的資料會被清空。
 
 2. 設定 **docker/.env** 目錄中的 `DOC_ENGINE` 為 `infinity`.
 
 3. 啟動容器:
 
    ```bash
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ docker compose -f docker-compose.yml up -d
    ```
 
 > [!WARNING]
@@ -265,7 +268,7 @@ docker build --build-arg NEED_MIRROR=1 -f Dockerfile -t infiniflow/ragflow:night
 
    ```bash
    pipx install uv
-   export UV_INDEX=https://pypi.tuna.tsinghua.edu.cn/simple
+   export UV_INDEX=https://mirrors.aliyun.com/pypi/simple
    ```
 
 2. 下載原始碼並安裝 Python 依賴：

README_zh.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.16.0-brightgreen" alt="docker pull infiniflow/ragflow:v0.16.0">
+        <img src="https://img.shields.io/badge/docker_pull-ragflow:v0.17.2-brightgreen" alt="docker pull infiniflow/ragflow:v0.17.2">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -55,9 +55,10 @@
 
 ## 🔥 近期更新
 
+- 2025-02-28 结合互联网搜索（Tavily），对于任意大模型实现类似 Deep Research 的推理功能.
 - 2025-02-05 更新硅基流动的模型列表，增加了对 Deepseek-R1/DeepSeek-V3 的支持。
 - 2025-01-26 优化知识图谱的提取和应用，提供了多种配置选择。
-- 2024-12-18 升级了 Deepdoc 的文档布局分析模型。
+- 2024-12-18 升级了 DeepDoc 的文档布局分析模型。
 - 2024-12-04 支持知识库的 Pagerank 分数。
 - 2024-11-22 完善了 Agent 中的变量定义和使用。
 - 2024-11-01 对解析后的 chunk 加入关键词抽取和相关问题生成以提高召回的准确度。
@@ -146,19 +147,23 @@
 
 3. 进入 **docker** 文件夹，利用提前编译好的 Docker 镜像启动服务器：
 
-   > 运行以下命令会自动下载 RAGFlow slim Docker 镜像 `v0.16.0-slim`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.16.0-slim` 的 Docker 镜像，请在运行 `docker compose` 启动服务之前先更新 **docker/.env** 文件内的 `RAGFLOW_IMAGE` 变量。比如，你可以通过设置 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0` 来下载 RAGFlow 镜像的 `v0.16.0` 完整发行版。
+> [!CAUTION]
+> 请注意，目前官方提供的所有 Docker 镜像均基于 x86 架构构建，并不提供基于 ARM64 的 Docker 镜像。
+> 如果你的操作系统是 ARM64 架构，请参考[这篇文档](https://ragflow.io/docs/dev/build_docker_image)自行构建 Docker 镜像。
+
+   > 运行以下命令会自动下载 RAGFlow slim Docker 镜像 `v0.17.2-slim`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.17.2-slim` 的 Docker 镜像，请在运行 `docker compose` 启动服务之前先更新 **docker/.env** 文件内的 `RAGFLOW_IMAGE` 变量。比如，你可以通过设置 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2` 来下载 RAGFlow 镜像的 `v0.17.2` 完整发行版。
 
    ```bash
-   $ cd ragflow
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ cd ragflow/docker
+   $ docker compose -f docker-compose.yml up -d
    ```
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.16.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.16.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.17.2           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.17.2-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
-   | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
+   | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 
    > [!TIP]
    > 如果你遇到 Docker 镜像拉不下来的问题，可以在 **docker/.env** 文件内根据变量 `RAGFLOW_IMAGE` 的注释提示选择华为云或者阿里云的相应镜像。
@@ -182,12 +187,9 @@
     /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     * Running on all addresses (0.0.0.0)
-    * Running on http://127.0.0.1:9380
-    * Running on http://x.x.x.x:9380
-    INFO:werkzeug:Press CTRL+C to quit
    ```
 
-   > 如果您跳过这一步系统确认步骤就登录 RAGFlow，你的浏览器有可能会提示 `network anormal` 或 `网络异常`，因为 RAGFlow 可能并未完全启动成功。
+   > 如果您在没有看到上面的提示信息出来之前，就尝试登录 RAGFlow，你的浏览器有可能会提示 `network anormal` 或 `网络异常`。
 
 5. 在你的浏览器中输入你的服务器对应的 IP 地址并登录 RAGFlow。
    > 上面这个例子中，您只需输入 http://IP_OF_YOUR_MACHINE 即可：未改动过配置则无需输入端口（默认的 HTTP 服务端口 80）。
@@ -216,7 +218,7 @@
 > 所有系统配置都需要通过系统重启生效：
 >
 > ```bash
-> $ docker compose -f docker/docker-compose.yml up -d
+> $ docker compose -f docker-compose.yml up -d
 > ```
 
 ### 把文档引擎从 Elasticsearch 切换成为 Infinity
@@ -228,13 +230,14 @@ RAGFlow 默认使用 Elasticsearch 存储文本和向量数据. 如果要切换
    ```bash
    $ docker compose -f docker/docker-compose.yml down -v
    ```
+   Note: `-v` 将会删除 docker 容器的 volumes，已有的数据会被清空。
 
 2. 设置 **docker/.env** 目录中的 `DOC_ENGINE` 为 `infinity`.
 
 3. 启动容器:
 
    ```bash
-   $ docker compose -f docker/docker-compose.yml up -d
+   $ docker compose -f docker-compose.yml up -d
    ```
 
 > [!WARNING]
@@ -266,7 +269,7 @@ docker build --build-arg NEED_MIRROR=1 -f Dockerfile -t infiniflow/ragflow:night
 
    ```bash
    pipx install uv
-   export UV_INDEX=https://pypi.tuna.tsinghua.edu.cn/simple
+   export UV_INDEX=https://mirrors.aliyun.com/pypi/simple
    ```
 
 2. 下载源代码并安装 Python 依赖：

agent/canvas.py

@@ -15,7 +15,6 @@
 #
 import logging
 import json
-from abc import ABC
 from copy import deepcopy
 from functools import partial
 
@@ -25,7 +24,7 @@ from agent.component import component_class
 from agent.component.base import ComponentBase
 
 
-class Canvas(ABC):
+class Canvas:
     """
     dsl = {
         "components": {
@@ -162,7 +161,7 @@ class Canvas(ABC):
             self.components[k]["obj"].reset()
         self._embed_id = ""
 
-    def get_compnent_name(self, cid):
+    def get_component_name(self, cid):
         for n in self.dsl["graph"]["nodes"]:
             if cid == n["id"]:
                 return n["data"]["name"]
@@ -210,7 +209,7 @@ class Canvas(ABC):
                             if c not in waiting:
                                 waiting.append(c)
                             continue
-                    yield "*'{}'* is running...🕞".format(self.get_compnent_name(c))
+                    yield "*'{}'* is running...🕞".format(self.get_component_name(c))
 
                     if cpn.component_name.lower() == "iteration":
                         st_cpn = cpn.get_start()

agent/component/base.py

@@ -555,7 +555,7 @@ class ComponentBase(ABC):
                     eles.extend(self._canvas.get_component(cpn_id)["obj"]._param.query)
                     continue
 
-                eles.append({"name": self._canvas.get_compnent_name(cpn_id), "key": cpn_id})
+                eles.append({"name": self._canvas.get_component_name(cpn_id), "key": cpn_id})
             else:
                 eles.append({"key": q["value"], "name": q["value"], "value": q["value"]})
         return eles

agent/component/categorize.py

@@ -80,6 +80,7 @@ class Categorize(Generate, ABC):
 
     def _run(self, history, **kwargs):
         input = self.get_input()
+        input = " - ".join(input["content"]) if "content" in input else ""
         chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
         ans = chat_mdl.chat(self._param.get_prompt(input), [{"role": "user", "content": "\nCategory: "}],
                             self._param.gen_conf())
@@ -93,5 +94,5 @@ class Categorize(Generate, ABC):
     def debug(self, **kwargs):
         df = self._run([], **kwargs)
         cpn_id = df.iloc[0, 0]
-        return Categorize.be_output(self._canvas.get_compnent_name(cpn_id))
+        return Categorize.be_output(self._canvas.get_component_name(cpn_id))
 

agent/component/exesql.py

@@ -52,15 +52,16 @@ class ExeSQLParam(GenerateParam):
         self.check_positive_integer(self.top_n, "Number of records")
         if self.database == "rag_flow":
             if self.host == "ragflow-mysql":
-                raise ValueError("The host is not accessible.")
+                raise ValueError("For the security reason, it dose not support database named rag_flow.")
             if self.password == "infini_rag_flow":
-                raise ValueError("The host is not accessible.")
+                raise ValueError("For the security reason, it dose not support database named rag_flow.")
 
 
 class ExeSQL(Generate, ABC):
     component_name = "ExeSQL"
 
-    def _refactor(self,ans):
+    def _refactor(self, ans):
+        ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
         match = re.search(r"```sql\s*(.*?)\s*```", ans, re.DOTALL)
         if match:
             ans = match.group(1)  # Query content
@@ -78,7 +79,6 @@ class ExeSQL(Generate, ABC):
         ans = self.get_input()
         ans = "".join([str(a) for a in ans["content"]]) if "content" in ans else ""
         ans = self._refactor(ans)
-        logging.info("db_type: ",self._param.db_type)
         if self._param.db_type in ["mysql", "mariadb"]:
             db = pymysql.connect(db=self._param.database, user=self._param.username, host=self._param.host,
                                  port=self._param.port, password=self._param.password)
@@ -87,11 +87,11 @@ class ExeSQL(Generate, ABC):
                                   port=self._param.port, password=self._param.password)
         elif self._param.db_type == 'mssql':
             conn_str = (
-                r'DRIVER={ODBC Driver 17 for SQL Server};'
-                r'SERVER=' + self._param.host + ',' + str(self._param.port) + ';'
-                r'DATABASE=' + self._param.database + ';'
-                r'UID=' + self._param.username + ';'
-                r'PWD=' + self._param.password
+                    r'DRIVER={ODBC Driver 17 for SQL Server};'
+                    r'SERVER=' + self._param.host + ',' + str(self._param.port) + ';'
+                    r'DATABASE=' + self._param.database + ';'
+                    r'UID=' + self._param.username + ';'
+                    r'PWD=' + self._param.password
             )
             db = pyodbc.connect(conn_str)
         try:
@@ -101,51 +101,50 @@ class ExeSQL(Generate, ABC):
         if not hasattr(self, "_loop"):
             setattr(self, "_loop", 0)
             self._loop += 1
-        input_list=re.split(r';', ans.replace(r"\n", " "))
+        input_list = re.split(r';', ans.replace(r"\n", " "))
         sql_res = []
         for i in range(len(input_list)):
-            single_sql=input_list[i]
+            single_sql = input_list[i]
             while self._loop <= self._param.loop:
-                self._loop+=1
+                self._loop += 1
                 if not single_sql:
                     break
                 try:
-                    logging.info("single_sql: ", single_sql)
                     cursor.execute(single_sql)
                     if cursor.rowcount == 0:
                         sql_res.append({"content": "No record in the database!"})
                         break
                     if self._param.db_type == 'mssql':
-                        single_res  = pd.DataFrame.from_records(cursor.fetchmany(self._param.top_n),columns = [desc[0] for desc in cursor.description])
+                        single_res = pd.DataFrame.from_records(cursor.fetchmany(self._param.top_n),
+                                                               columns=[desc[0] for desc in cursor.description])
                     else:
                         single_res = pd.DataFrame([i for i in cursor.fetchmany(self._param.top_n)])
                         single_res.columns = [i[0] for i in cursor.description]
-                    sql_res.append({"content":  single_res.to_markdown()})
+                    sql_res.append({"content": single_res.to_markdown(index=False, floatfmt=".6f")})
                     break
                 except Exception as e:
                     single_sql = self._regenerate_sql(single_sql, str(e), **kwargs)
                     single_sql = self._refactor(single_sql)
                     if self._loop > self._param.loop:
                         sql_res.append({"content": "Can't query the correct data via SQL statement."})
-                        # raise Exception("Maximum loop time exceeds. Can't query the correct data via SQL statement.")
         db.close()
         if not sql_res:
             return ExeSQL.be_output("")
         return pd.DataFrame(sql_res)
 
-    def _regenerate_sql(self, failed_sql, error_message,**kwargs):
+    def _regenerate_sql(self, failed_sql, error_message, **kwargs):
         prompt = f'''
         ## You are the Repair SQL Statement Helper, please modify the original SQL statement based on the SQL query error report.
         ## The original SQL statement is as follows:{failed_sql}.
         ## The contents of the SQL query error report is as follows:{error_message}.
         ## Answer only the modified SQL statement. Please do not give any explanation, just answer the code.
 '''
-        self._param.prompt=prompt
+        self._param.prompt = prompt
         kwargs_ = deepcopy(kwargs)
         kwargs_["stream"] = False
         response = Generate._run(self, [], **kwargs_)
         try:
-            regenerated_sql = response.loc[0,"content"]
+            regenerated_sql = response.loc[0, "content"]
             return regenerated_sql
         except Exception as e:
             logging.error(f"Failed to regenerate SQL: {e}")

agent/component/generate.py

@@ -18,10 +18,10 @@ from functools import partial
 import pandas as pd
 from api.db import LLMType
 from api.db.services.conversation_service import structure_answer
-from api.db.services.dialog_service import message_fit_in
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from agent.component.base import ComponentBase, ComponentParamBase
+from rag.prompts import message_fit_in
 
 
 class GenerateParam(ComponentParamBase):
@@ -69,10 +69,8 @@ class Generate(ComponentBase):
     component_name = "Generate"
 
     def get_dependent_components(self):
-        cpnts = set([para["component_id"].split("@")[0] for para in self._param.parameters \
-                     if para.get("component_id") \
-                     and para["component_id"].lower().find("answer") < 0 \
-                     and para["component_id"].lower().find("begin") < 0])
+        inputs = self.get_input_elements()
+        cpnts = set([i["key"] for i in inputs[1:] if i["key"].lower().find("answer") < 0 and i["key"].lower().find("begin") < 0])
         return list(cpnts)
 
     def set_cite(self, retrieval_res, answer):
@@ -110,10 +108,26 @@ class Generate(ComponentBase):
         return res
 
     def get_input_elements(self):
-        if self._param.parameters:
-            return [{"key": "user", "name": "Input your question here:"}, *self._param.parameters]
-
-        return [{"key": "user", "name": "Input your question here:"}]
+        key_set = set([])
+        res = [{"key": "user", "name": "Input your question here:"}]
+        for r in re.finditer(r"\{([a-z]+[:@][a-z0-9_-]+)\}", self._param.prompt, flags=re.IGNORECASE):
+            cpn_id = r.group(1)
+            if cpn_id in key_set:
+                continue
+            if cpn_id.lower().find("begin@") == 0:
+                cpn_id, key = cpn_id.split("@")
+                for p in self._canvas.get_component(cpn_id)["obj"]._param.query:
+                    if p["key"] != key:
+                        continue
+                    res.append({"key": r.group(1), "name": p["name"]})
+                    key_set.add(r.group(1))
+                continue
+            cpn_nm = self._canvas.get_component_name(cpn_id)
+            if not cpn_nm:
+                continue
+            res.append({"key": cpn_id, "name": cpn_nm})
+            key_set.add(cpn_id)
+        return res
 
     def _run(self, history, **kwargs):
         chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
@@ -121,22 +135,20 @@ class Generate(ComponentBase):
 
         retrieval_res = []
         self._param.inputs = []
-        for para in self._param.parameters:
-            if not para.get("component_id"):
-                continue
-            component_id = para["component_id"].split("@")[0]
-            if para["component_id"].lower().find("@") >= 0:
-                cpn_id, key = para["component_id"].split("@")
+        for para in self.get_input_elements()[1:]:
+            if para["key"].lower().find("begin@") == 0:
+                cpn_id, key = para["key"].split("@")
                 for p in self._canvas.get_component(cpn_id)["obj"]._param.query:
                     if p["key"] == key:
                         kwargs[para["key"]] = p.get("value", "")
                         self._param.inputs.append(
-                            {"component_id": para["component_id"], "content": kwargs[para["key"]]})
+                            {"component_id": para["key"], "content": kwargs[para["key"]]})
                         break
                 else:
                     assert False, f"Can't find parameter '{key}' for {cpn_id}"
                 continue
 
+            component_id = para["key"]
             cpn = self._canvas.get_component(component_id)["obj"]
             if cpn.component_name.lower() == "answer":
                 hist = self._canvas.get_history(1)
@@ -152,8 +164,8 @@ class Generate(ComponentBase):
             else:
                 if cpn.component_name.lower() == "retrieval":
                     retrieval_res.append(out)
-                kwargs[para["key"]] = "  - "+"\n - ".join([o if isinstance(o, str) else str(o) for o in out["content"]])
-            self._param.inputs.append({"component_id": para["component_id"], "content": kwargs[para["key"]]})
+                kwargs[para["key"]] = "  - " + "\n - ".join([o if isinstance(o, str) else str(o) for o in out["content"]])
+            self._param.inputs.append({"component_id": para["key"], "content": kwargs[para["key"]]})
 
         if retrieval_res:
             retrieval_res = pd.concat(retrieval_res, ignore_index=True)
@@ -175,17 +187,18 @@ class Generate(ComponentBase):
             return partial(self.stream_output, chat_mdl, prompt, retrieval_res)
 
         if "empty_response" in retrieval_res.columns and not "".join(retrieval_res["content"]):
-            res = {"content": "\n- ".join(retrieval_res["empty_response"]) if "\n- ".join(
-                retrieval_res["empty_response"]) else "Nothing found in knowledgebase!", "reference": []}
+            empty_res = "\n- ".join([str(t) for t in retrieval_res["empty_response"] if str(t)])
+            res = {"content": empty_res if empty_res else "Nothing found in knowledgebase!", "reference": []}
             return pd.DataFrame([res])
 
         msg = self._canvas.get_history(self._param.message_history_window_size)
         if len(msg) < 1:
-            msg.append({"role": "user", "content": ""})
+            msg.append({"role": "user", "content": "Output: "})
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97))
         if len(msg) < 2:
-            msg.append({"role": "user", "content": ""})
+            msg.append({"role": "user", "content": "Output: "})
         ans = chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf())
+        ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
 
         if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
             res = self.set_cite(retrieval_res, ans)
@@ -196,18 +209,20 @@ class Generate(ComponentBase):
     def stream_output(self, chat_mdl, prompt, retrieval_res):
         res = None
         if "empty_response" in retrieval_res.columns and not "".join(retrieval_res["content"]):
-            res = {"content": "\n- ".join(retrieval_res["empty_response"]) if "\n- ".join(
-                retrieval_res["empty_response"]) else "Nothing found in knowledgebase!", "reference": []}
+            empty_res = "\n- ".join([str(t) for t in retrieval_res["empty_response"] if str(t)])
+            res = {"content": empty_res if empty_res else "Nothing found in knowledgebase!", "reference": []}
             yield res
             self.set_output(res)
             return
 
         msg = self._canvas.get_history(self._param.message_history_window_size)
+        if msg and msg[0]['role'] == 'assistant':
+            msg.pop(0)
         if len(msg) < 1:
-            msg.append({"role": "user", "content": ""})
+            msg.append({"role": "user", "content": "Output: "})
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97))
         if len(msg) < 2:
-            msg.append({"role": "user", "content": ""})
+            msg.append({"role": "user", "content": "Output: "})
         answer = ""
         for ans in chat_mdl.chat_streamly(msg[0]["content"], msg[1:], self._param.gen_conf()):
             res = {"content": ans, "reference": []}
@@ -230,5 +245,6 @@ class Generate(ComponentBase):
         for n, v in kwargs.items():
             prompt = re.sub(r"\{%s\}" % re.escape(n), str(v).replace("\\", " "), prompt)
 
-        ans = chat_mdl.chat(prompt, [{"role": "user", "content": kwargs.get("user", "")}], self._param.gen_conf())
+        u = kwargs.get("user")
+        ans = chat_mdl.chat(prompt, [{"role": "user", "content": u if u else "Output: "}], self._param.gen_conf())
         return pd.DataFrame([ans])

agent/component/invoke.py

@@ -35,12 +35,14 @@ class InvokeParam(ComponentParamBase):
         self.url = ""
         self.timeout = 60
         self.clean_html = False
+        self.datatype = "json"  # New parameter to determine data posting type
 
     def check(self):
         self.check_valid_value(self.method.lower(), "Type of content from the crawler", ['get', 'post', 'put'])
         self.check_empty(self.url, "End point URL")
         self.check_positive_integer(self.timeout, "Timeout time in second")
         self.check_boolean(self.clean_html, "Clean HTML")
+        self.check_valid_value(self.datatype.lower(), "Data post type", ['json', 'formdata'])  # Check for valid datapost value
 
 
 class Invoke(ComponentBase, ABC):
@@ -94,22 +96,36 @@ class Invoke(ComponentBase, ABC):
             return Invoke.be_output(response.text)
 
         if method == 'put':
-            response = requests.put(url=url,
-                                    data=args,
-                                    headers=headers,
-                                    proxies=proxies,
-                                    timeout=self._param.timeout)
+            if self._param.datatype.lower() == 'json':
+                response = requests.put(url=url,
+                                        json=args,
+                                        headers=headers,
+                                        proxies=proxies,
+                                        timeout=self._param.timeout)
+            else:
+                response = requests.put(url=url,
+                                        data=args,
+                                        headers=headers,
+                                        proxies=proxies,
+                                        timeout=self._param.timeout)
             if self._param.clean_html:
                 sections = HtmlParser()(None, response.content)
                 return Invoke.be_output("\n".join(sections))
             return Invoke.be_output(response.text)
 
         if method == 'post':
-            response = requests.post(url=url,
-                                    json=args,
-                                    headers=headers,
-                                    proxies=proxies,
-                                    timeout=self._param.timeout)
+            if self._param.datatype.lower() == 'json':
+                response = requests.post(url=url,
+                                         json=args,
+                                         headers=headers,
+                                         proxies=proxies,
+                                         timeout=self._param.timeout)
+            else:
+                response = requests.post(url=url,
+                                         data=args,
+                                         headers=headers,
+                                         proxies=proxies,
+                                         timeout=self._param.timeout)
             if self._param.clean_html:
                 sections = HtmlParser()(None, response.content)
                 return Invoke.be_output("\n".join(sections))

agent/component/iterationitem.py

@@ -38,6 +38,10 @@ class IterationItem(ComponentBase, ABC):
         ans = parent.get_input()
         ans = parent._param.delimiter.join(ans["content"]) if "content" in ans else ""
         ans = [a.strip() for a in ans.split(parent._param.delimiter)]
+        if not ans:
+            self._idx = -1
+            return pd.DataFrame()
+
         df = pd.DataFrame([{"content": ans[self._idx]}])
         self._idx += 1
         if self._idx >= len(ans):

agent/component/retrieval.py

@@ -19,11 +19,12 @@ from abc import ABC
 import pandas as pd
 
 from api.db import LLMType
-from api.db.services.dialog_service import label_question
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from agent.component.base import ComponentBase, ComponentParamBase
+from rag.app.tag import label_question
+from rag.utils.tavily_conn import Tavily
 
 
 class RetrievalParam(ComponentParamBase):
@@ -40,10 +41,12 @@ class RetrievalParam(ComponentParamBase):
         self.kb_ids = []
         self.rerank_id = ""
         self.empty_response = ""
+        self.tavily_api_key = ""
+        self.use_kg = False
 
     def check(self):
         self.check_decimal_float(self.similarity_threshold, "[Retrieval] Similarity threshold")
-        self.check_decimal_float(self.keywords_similarity_weight, "[Retrieval] Keywords similarity weight")
+        self.check_decimal_float(self.keywords_similarity_weight, "[Retrieval] Keyword similarity weight")
         self.check_positive_number(self.top_n, "[Retrieval] Top N")
 
 
@@ -53,7 +56,9 @@ class Retrieval(ComponentBase, ABC):
     def _run(self, history, **kwargs):
         query = self.get_input()
         query = str(query["content"][0]) if "content" in query else ""
-
+        lines = query.split('\n')
+        user_queries = [line.split("USER:", 1)[1] for line in lines if line.startswith("USER:")]
+        query = user_queries[-1] if user_queries else ""
         kbs = KnowledgebaseService.get_by_ids(self._param.kb_ids)
         if not kbs:
             return Retrieval.be_output("")
@@ -73,6 +78,20 @@ class Retrieval(ComponentBase, ABC):
                                         self._param.similarity_threshold, 1 - self._param.keywords_similarity_weight,
                                         aggs=False, rerank_mdl=rerank_mdl,
                                         rank_feature=label_question(query, kbs))
+        if self._param.use_kg:
+            ck = settings.kg_retrievaler.retrieval(query,
+                                                   [kbs[0].tenant_id],
+                                                   self._param.kb_ids,
+                                                   embd_mdl,
+                                                   LLMBundle(kbs[0].tenant_id, LLMType.CHAT))
+            if ck["content_with_weight"]:
+                kbinfos["chunks"].insert(0, ck)
+
+        if self._param.tavily_api_key:
+            tav = Tavily(self._param.tavily_api_key)
+            tav_res = tav.retrieve_chunks(query)
+            kbinfos["chunks"].extend(tav_res["chunks"])
+            kbinfos["doc_aggs"].extend(tav_res["doc_aggs"])
 
         if not kbinfos["chunks"]:
             df = Retrieval.be_output("")

agent/component/rewrite.py

@@ -13,93 +13,82 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
-import logging
 from abc import ABC
-from api.db import LLMType
-from api.db.services.llm_service import LLMBundle
 from agent.component import GenerateParam, Generate
+from rag.prompts import full_question
 
 
 class RewriteQuestionParam(GenerateParam):
-
     """
     Define the QuestionRewrite component parameters.
     """
+
     def __init__(self):
         super().__init__()
         self.temperature = 0.9
         self.prompt = ""
+        self.language = ""
 
     def check(self):
         super().check()
 
-    def get_prompt(self, conv):
-        self.prompt = """
-        You are an expert at query expansion to generate a paraphrasing of a question.
-        I can't retrieval relevant information from the knowledge base by using user's question directly.     
-        You need to expand or paraphrase user's question by multiple ways such as using synonyms words/phrase, 
-        writing the abbreviation in its entirety, adding some extra descriptions or explanations, 
-        changing the way of expression, translating the original question into another language (English/Chinese), etc. 
-        And return 5 versions of question and one is from translation.
-        Just list the question. No other words are needed.
-        """
-        return f"""
-Role: A helpful assistant
-Task: Generate a full user question that would follow the conversation.
-Requirements & Restrictions:
-  - Text generated MUST be in the same language of the original user's question.
-  - If the user's latest question is completely, don't do anything, just return the original question.
-  - DON'T generate anything except a refined question.
-
-######################
--Examples-
-######################
-# Example 1
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-###############
-Output: What's the name of Donald Trump's mother?
-------------
-# Example 2
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-ASSISTANT:  Mary Trump.
-User: What's her full name?
-###############
-Output: What's the full name of Donald Trump's mother Mary Trump?
-######################
-# Real Data
-## Conversation
-{conv}
-###############
-    """
-        return self.prompt
-
 
 class RewriteQuestion(Generate, ABC):
     component_name = "RewriteQuestion"
 
     def _run(self, history, **kwargs):
         hist = self._canvas.get_history(self._param.message_history_window_size)
-        conv = []
-        for m in hist:
-            if m["role"] not in ["user", "assistant"]:
-                continue
-            conv.append("{}: {}".format(m["role"].upper(), m["content"]))
-        conv = "\n".join(conv)
-
-        chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
-        ans = chat_mdl.chat(self._param.get_prompt(conv), [{"role": "user", "content": "Output: "}],
-                            self._param.gen_conf())
+        query = self.get_input()
+        query = str(query["content"][0]) if "content" in query else ""
+        messages = [h for h in hist if h["role"]!="system"]
+        if messages[-1]["role"] != "user":
+            messages.append({"role": "user", "content": query})
+        ans = full_question(self._canvas.get_tenant_id(), self._param.llm_id, messages, self.gen_lang(self._param.language))
         self._canvas.history.pop()
         self._canvas.history.append(("user", ans))
-
-        logging.debug(ans)
         return RewriteQuestion.be_output(ans)
 
-
-
+    @staticmethod
+    def gen_lang(language):
+        # convert code lang to language word for the prompt
+        language_dict = {'af': 'Afrikaans', 'ak': 'Akan', 'sq': 'Albanian', 'ws': 'Samoan', 'am': 'Amharic',
+                         'ar': 'Arabic', 'hy': 'Armenian', 'az': 'Azerbaijani', 'eu': 'Basque', 'be': 'Belarusian',
+                         'bem': 'Bemba', 'bn': 'Bengali', 'bh': 'Bihari',
+                         'xx-bork': 'Bork', 'bs': 'Bosnian', 'br': 'Breton', 'bg': 'Bulgarian', 'bt': 'Bhutani',
+                         'km': 'Cambodian', 'ca': 'Catalan', 'chr': 'Cherokee', 'ny': 'Chichewa', 'zh-cn': 'Chinese',
+                         'zh-tw': 'Chinese', 'co': 'Corsican',
+                         'hr': 'Croatian', 'cs': 'Czech', 'da': 'Danish', 'nl': 'Dutch', 'xx-elmer': 'Elmer',
+                         'en': 'English', 'eo': 'Esperanto', 'et': 'Estonian', 'ee': 'Ewe', 'fo': 'Faroese',
+                         'tl': 'Filipino', 'fi': 'Finnish', 'fr': 'French',
+                         'fy': 'Frisian', 'gaa': 'Ga', 'gl': 'Galician', 'ka': 'Georgian', 'de': 'German',
+                         'el': 'Greek', 'kl': 'Greenlandic', 'gn': 'Guarani', 'gu': 'Gujarati', 'xx-hacker': 'Hacker',
+                         'ht': 'Haitian Creole', 'ha': 'Hausa', 'haw': 'Hawaiian',
+                         'iw': 'Hebrew', 'hi': 'Hindi', 'hu': 'Hungarian', 'is': 'Icelandic', 'ig': 'Igbo',
+                         'id': 'Indonesian', 'ia': 'Interlingua', 'ga': 'Irish', 'it': 'Italian', 'ja': 'Japanese',
+                         'jw': 'Javanese', 'kn': 'Kannada', 'kk': 'Kazakh', 'rw': 'Kinyarwanda',
+                         'rn': 'Kirundi', 'xx-klingon': 'Klingon', 'kg': 'Kongo', 'ko': 'Korean', 'kri': 'Krio',
+                         'ku': 'Kurdish', 'ckb': 'Kurdish (Sorani)', 'ky': 'Kyrgyz', 'lo': 'Laothian', 'la': 'Latin',
+                         'lv': 'Latvian', 'ln': 'Lingala', 'lt': 'Lithuanian',
+                         'loz': 'Lozi', 'lg': 'Luganda', 'ach': 'Luo', 'mk': 'Macedonian', 'mg': 'Malagasy',
+                         'ms': 'Malay', 'ml': 'Malayalam', 'mt': 'Maltese', 'mv': 'Maldivian', 'mi': 'Maori',
+                         'mr': 'Marathi', 'mfe': 'Mauritian Creole', 'mo': 'Moldavian', 'mn': 'Mongolian',
+                         'sr-me': 'Montenegrin', 'my': 'Burmese', 'ne': 'Nepali', 'pcm': 'Nigerian Pidgin',
+                         'nso': 'Northern Sotho', 'no': 'Norwegian', 'nn': 'Norwegian Nynorsk', 'oc': 'Occitan',
+                         'or': 'Oriya', 'om': 'Oromo', 'ps': 'Pashto', 'fa': 'Persian',
+                         'xx-pirate': 'Pirate', 'pl': 'Polish', 'pt': 'Portuguese', 'pt-br': 'Portuguese (Brazilian)',
+                         'pt-pt': 'Portuguese (Portugal)', 'pa': 'Punjabi', 'qu': 'Quechua', 'ro': 'Romanian',
+                         'rm': 'Romansh', 'nyn': 'Runyankole', 'ru': 'Russian', 'gd': 'Scots Gaelic',
+                         'sr': 'Serbian', 'sh': 'Serbo-Croatian', 'st': 'Sesotho', 'tn': 'Setswana',
+                         'crs': 'Seychellois Creole', 'sn': 'Shona', 'sd': 'Sindhi', 'si': 'Sinhalese', 'sk': 'Slovak',
+                         'sl': 'Slovenian', 'so': 'Somali', 'es': 'Spanish', 'es-419': 'Spanish (Latin America)',
+                         'su': 'Sundanese',
+                         'sw': 'Swahili', 'sv': 'Swedish', 'tg': 'Tajik', 'ta': 'Tamil', 'tt': 'Tatar', 'te': 'Telugu',
+                         'th': 'Thai', 'ti': 'Tigrinya', 'to': 'Tongan', 'lua': 'Tshiluba', 'tum': 'Tumbuka',
+                         'tr': 'Turkish', 'tk': 'Turkmen', 'tw': 'Twi',
+                         'ug': 'Uyghur', 'uk': 'Ukrainian', 'ur': 'Urdu', 'uz': 'Uzbek', 'vu': 'Vanuatu',
+                         'vi': 'Vietnamese', 'cy': 'Welsh', 'wo': 'Wolof', 'xh': 'Xhosa', 'yi': 'Yiddish',
+                         'yo': 'Yoruba', 'zu': 'Zulu'}
+        if language in language_dict:
+            return language_dict[language]
+        else:
+            return ""

agent/component/template.py

@@ -38,27 +38,39 @@ class Template(ComponentBase):
     component_name = "Template"
 
     def get_dependent_components(self):
-        cpnts = set(
-            [
-                para["component_id"].split("@")[0]
-                for para in self._param.parameters
-                if para.get("component_id")
-                and para["component_id"].lower().find("answer") < 0
-                and para["component_id"].lower().find("begin") < 0
-            ]
-        )
+        inputs = self.get_input_elements()
+        cpnts = set([i["key"] for i in inputs if i["key"].lower().find("answer") < 0 and i["key"].lower().find("begin") < 0])
         return list(cpnts)
 
+    def get_input_elements(self):
+        key_set = set([])
+        res = []
+        for r in re.finditer(r"\{([a-z]+[:@][a-z0-9_-]+)\}", self._param.content, flags=re.IGNORECASE):
+            cpn_id = r.group(1)
+            if cpn_id in key_set:
+                continue
+            if cpn_id.lower().find("begin@") == 0:
+                cpn_id, key = cpn_id.split("@")
+                for p in self._canvas.get_component(cpn_id)["obj"]._param.query:
+                    if p["key"] != key:
+                        continue
+                    res.append({"key": r.group(1), "name": p["name"]})
+                    key_set.add(r.group(1))
+                continue
+            cpn_nm = self._canvas.get_component_name(cpn_id)
+            if not cpn_nm:
+                continue
+            res.append({"key": cpn_id, "name": cpn_nm})
+            key_set.add(cpn_id)
+        return res
+
     def _run(self, history, **kwargs):
         content = self._param.content
 
         self._param.inputs = []
-        for para in self._param.parameters:
-            if not para.get("component_id"):
-                continue
-            component_id = para["component_id"].split("@")[0]
-            if para["component_id"].lower().find("@") >= 0:
-                cpn_id, key = para["component_id"].split("@")
+        for para in self.get_input_elements():
+            if para["key"].lower().find("begin@") == 0:
+                cpn_id, key = para["key"].split("@")
                 for p in self._canvas.get_component(cpn_id)["obj"]._param.query:
                     if p["key"] == key:
                         value = p.get("value", "")
@@ -68,6 +80,7 @@ class Template(ComponentBase):
                     assert False, f"Can't find parameter '{key}' for {cpn_id}"
                 continue
 
+            component_id = para["key"]
             cpn = self._canvas.get_component(component_id)["obj"]
             if cpn.component_name.lower() == "answer":
                 hist = self._canvas.get_history(1)
@@ -114,7 +127,7 @@ class Template(ComponentBase):
 
     def make_kwargs(self, para, kwargs, value):
         self._param.inputs.append(
-            {"component_id": para["component_id"], "content": value}
+            {"component_id": para["key"], "content": value}
         )
         try:
             value = json.loads(value)

agentic_reasoning/__init__.py

@@ -0,0 +1 @@
+from .deep_research import DeepResearcher as DeepResearcher

agentic_reasoning/deep_research.py

@@ -0,0 +1,223 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import logging
+import re
+from functools import partial
+from agentic_reasoning.prompts import BEGIN_SEARCH_QUERY, BEGIN_SEARCH_RESULT, END_SEARCH_RESULT, MAX_SEARCH_LIMIT, \
+    END_SEARCH_QUERY, REASON_PROMPT, RELEVANT_EXTRACTION_PROMPT
+from api.db.services.llm_service import LLMBundle
+from rag.nlp import extract_between
+from rag.prompts import kb_prompt
+from rag.utils.tavily_conn import Tavily
+
+
+class DeepResearcher:
+    def __init__(self,
+                 chat_mdl: LLMBundle,
+                 prompt_config: dict,
+                 kb_retrieve: partial = None,
+                 kg_retrieve: partial = None
+                 ):
+        self.chat_mdl = chat_mdl
+        self.prompt_config = prompt_config
+        self._kb_retrieve = kb_retrieve
+        self._kg_retrieve = kg_retrieve
+
+    @staticmethod
+    def _remove_query_tags(text):
+        """Remove query tags from text"""
+        pattern = re.escape(BEGIN_SEARCH_QUERY) + r"(.*?)" + re.escape(END_SEARCH_QUERY)
+        return re.sub(pattern, "", text)
+
+    @staticmethod
+    def _remove_result_tags(text):
+        """Remove result tags from text"""
+        pattern = re.escape(BEGIN_SEARCH_RESULT) + r"(.*?)" + re.escape(END_SEARCH_RESULT)
+        return re.sub(pattern, "", text)
+
+    def _generate_reasoning(self, msg_history):
+        """Generate reasoning steps"""
+        query_think = ""
+        if msg_history[-1]["role"] != "user":
+            msg_history.append({"role": "user", "content": "Continues reasoning with the new information.\n"})
+        else:
+            msg_history[-1]["content"] += "\n\nContinues reasoning with the new information.\n"
+            
+        for ans in self.chat_mdl.chat_streamly(REASON_PROMPT, msg_history, {"temperature": 0.7}):
+            ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
+            if not ans:
+                continue
+            query_think = ans
+            yield query_think
+        return query_think
+
+    def _extract_search_queries(self, query_think, question, step_index):
+        """Extract search queries from thinking"""
+        queries = extract_between(query_think, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY)
+        if not queries and step_index == 0:
+            # If this is the first step and no queries are found, use the original question as the query
+            queries = [question]
+        return queries
+
+    def _truncate_previous_reasoning(self, all_reasoning_steps):
+        """Truncate previous reasoning steps to maintain a reasonable length"""
+        truncated_prev_reasoning = ""
+        for i, step in enumerate(all_reasoning_steps):
+            truncated_prev_reasoning += f"Step {i + 1}: {step}\n\n"
+
+        prev_steps = truncated_prev_reasoning.split('\n\n')
+        if len(prev_steps) <= 5:
+            truncated_prev_reasoning = '\n\n'.join(prev_steps)
+        else:
+            truncated_prev_reasoning = ''
+            for i, step in enumerate(prev_steps):
+                if i == 0 or i >= len(prev_steps) - 4 or BEGIN_SEARCH_QUERY in step or BEGIN_SEARCH_RESULT in step:
+                    truncated_prev_reasoning += step + '\n\n'
+                else:
+                    if truncated_prev_reasoning[-len('\n\n...\n\n'):] != '\n\n...\n\n':
+                        truncated_prev_reasoning += '...\n\n'
+        
+        return truncated_prev_reasoning.strip('\n')
+
+    def _retrieve_information(self, search_query):
+        """Retrieve information from different sources"""
+        # 1. Knowledge base retrieval
+        kbinfos = self._kb_retrieve(question=search_query) if self._kb_retrieve else {"chunks": [], "doc_aggs": []}
+        
+        # 2. Web retrieval (if Tavily API is configured)
+        if self.prompt_config.get("tavily_api_key"):
+            tav = Tavily(self.prompt_config["tavily_api_key"])
+            tav_res = tav.retrieve_chunks(search_query)
+            kbinfos["chunks"].extend(tav_res["chunks"])
+            kbinfos["doc_aggs"].extend(tav_res["doc_aggs"])
+        
+        # 3. Knowledge graph retrieval (if configured)
+        if self.prompt_config.get("use_kg") and self._kg_retrieve:
+            ck = self._kg_retrieve(question=search_query)
+            if ck["content_with_weight"]:
+                kbinfos["chunks"].insert(0, ck)
+                
+        return kbinfos
+
+    def _update_chunk_info(self, chunk_info, kbinfos):
+        """Update chunk information for citations"""
+        if not chunk_info["chunks"]:
+            # If this is the first retrieval, use the retrieval results directly
+            for k in chunk_info.keys():
+                chunk_info[k] = kbinfos[k]
+        else:
+            # Merge newly retrieved information, avoiding duplicates
+            cids = [c["chunk_id"] for c in chunk_info["chunks"]]
+            for c in kbinfos["chunks"]:
+                if c["chunk_id"] not in cids:
+                    chunk_info["chunks"].append(c)
+                    
+            dids = [d["doc_id"] for d in chunk_info["doc_aggs"]]
+            for d in kbinfos["doc_aggs"]:
+                if d["doc_id"] not in dids:
+                    chunk_info["doc_aggs"].append(d)
+
+    def _extract_relevant_info(self, truncated_prev_reasoning, search_query, kbinfos):
+        """Extract and summarize relevant information"""
+        summary_think = ""
+        for ans in self.chat_mdl.chat_streamly(
+                RELEVANT_EXTRACTION_PROMPT.format(
+                    prev_reasoning=truncated_prev_reasoning,
+                    search_query=search_query,
+                    document="\n".join(kb_prompt(kbinfos, 4096))
+                ),
+                [{"role": "user",
+                  "content": f'Now you should analyze each web page and find helpful information based on the current search query "{search_query}" and previous reasoning steps.'}],
+                {"temperature": 0.7}):
+            ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
+            if not ans:
+                continue
+            summary_think = ans
+            yield summary_think
+        
+        return summary_think
+
+    def thinking(self, chunk_info: dict, question: str):
+        executed_search_queries = []
+        msg_history = [{"role": "user", "content": f'Question:\"{question}\"\n'}]
+        all_reasoning_steps = []
+        think = "<think>"
+        
+        for step_index in range(MAX_SEARCH_LIMIT + 1):
+            # Check if the maximum search limit has been reached
+            if step_index == MAX_SEARCH_LIMIT - 1:
+                summary_think = f"\n{BEGIN_SEARCH_RESULT}\nThe maximum search limit is exceeded. You are not allowed to search.\n{END_SEARCH_RESULT}\n"
+                yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
+                all_reasoning_steps.append(summary_think)
+                msg_history.append({"role": "assistant", "content": summary_think})
+                break
+
+            # Step 1: Generate reasoning
+            query_think = ""
+            for ans in self._generate_reasoning(msg_history):
+                query_think = ans
+                yield {"answer": think + self._remove_query_tags(query_think) + "</think>", "reference": {}, "audio_binary": None}
+
+            think += self._remove_query_tags(query_think)
+            all_reasoning_steps.append(query_think)
+            
+            # Step 2: Extract search queries
+            queries = self._extract_search_queries(query_think, question, step_index)
+            if not queries and step_index > 0:
+                # If not the first step and no queries, end the search process
+                break
+
+            # Process each search query
+            for search_query in queries:
+                logging.info(f"[THINK]Query: {step_index}. {search_query}")
+                msg_history.append({"role": "assistant", "content": search_query})
+                think += f"\n\n> {step_index + 1}. {search_query}\n\n"
+                yield {"answer": think + "</think>", "reference": {}, "audio_binary": None}
+
+                # Check if the query has already been executed
+                if search_query in executed_search_queries:
+                    summary_think = f"\n{BEGIN_SEARCH_RESULT}\nYou have searched this query. Please refer to previous results.\n{END_SEARCH_RESULT}\n"
+                    yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
+                    all_reasoning_steps.append(summary_think)
+                    msg_history.append({"role": "user", "content": summary_think})
+                    think += summary_think
+                    continue
+                
+                executed_search_queries.append(search_query)
+                
+                # Step 3: Truncate previous reasoning steps
+                truncated_prev_reasoning = self._truncate_previous_reasoning(all_reasoning_steps)
+                
+                # Step 4: Retrieve information
+                kbinfos = self._retrieve_information(search_query)
+                
+                # Step 5: Update chunk information
+                self._update_chunk_info(chunk_info, kbinfos)
+                
+                # Step 6: Extract relevant information
+                think += "\n\n"
+                summary_think = ""
+                for ans in self._extract_relevant_info(truncated_prev_reasoning, search_query, kbinfos):
+                    summary_think = ans
+                    yield {"answer": think + self._remove_result_tags(summary_think) + "</think>", "reference": {}, "audio_binary": None}
+
+                all_reasoning_steps.append(summary_think)
+                msg_history.append(
+                    {"role": "user", "content": f"\n\n{BEGIN_SEARCH_RESULT}{summary_think}{END_SEARCH_RESULT}\n\n"})
+                think += self._remove_result_tags(summary_think)
+                logging.info(f"[THINK]Summary: {step_index}. {summary_think}")
+
+        yield think + "</think>"

agentic_reasoning/prompts.py

@@ -0,0 +1,113 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
+END_SEARCH_QUERY = "<|end_search_query|>"
+BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
+END_SEARCH_RESULT = "<|end_search_result|>"
+MAX_SEARCH_LIMIT = 6
+
+REASON_PROMPT = (
+        "You are a reasoning assistant with the ability to perform dataset searches to help "
+        "you answer the user's question accurately. You have special tools:\n\n"
+        f"- To perform a search: write {BEGIN_SEARCH_QUERY} your query here {END_SEARCH_QUERY}.\n"
+        f"Then, the system will search and analyze relevant content, then provide you with helpful information in the format {BEGIN_SEARCH_RESULT} ...search results... {END_SEARCH_RESULT}.\n\n"
+        f"You can repeat the search process multiple times if necessary. The maximum number of search attempts is limited to {MAX_SEARCH_LIMIT}.\n\n"
+        "Once you have all the information you need, continue your reasoning.\n\n"
+        "-- Example 1 --\n" ########################################
+        "Question: \"Are both the directors of Jaws and Casino Royale from the same country?\"\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who is the director of Jaws?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nThe director of Jaws is Steven Spielberg...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information.\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Where is Steven Spielberg from?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nSteven Allan Spielberg is an American filmmaker...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who is the director of Casino Royale?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCasino Royale is a 2006 spy film directed by Martin Campbell...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Where is Martin Campbell from?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nMartin Campbell (born 24 October 1943) is a New Zealand film and television director...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\nIt's enough to answer the question\n"
+
+        "-- Example 2 --\n" #########################################
+        "Question: \"When was the founder of craigslist born?\"\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who was the founder of craigslist?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCraigslist was founded by Craig Newmark...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information.\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY} When was Craig Newmark born?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCraig Newmark was born on December 6, 1952...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\nIt's enough to answer the question\n"
+        "**Remember**:\n"
+        f"- You have a dataset to search, so you just provide a proper search query.\n"
+        f"- Use {BEGIN_SEARCH_QUERY} to request a dataset search and end with {END_SEARCH_QUERY}.\n"
+        "- The language of query MUST be as the same as 'Question' or 'search result'.\n"
+        "- If no helpful information can be found, rewrite the search query to be less and precise keywords.\n"
+        "- When done searching, continue your reasoning.\n\n"
+        'Please answer the following question. You should think step by step to solve it.\n\n'
+    )
+
+RELEVANT_EXTRACTION_PROMPT = """**Task Instruction:**
+
+    You are tasked with reading and analyzing web pages based on the following inputs: **Previous Reasoning Steps**, **Current Search Query**, and **Searched Web Pages**. Your objective is to extract relevant and helpful information for **Current Search Query** from the **Searched Web Pages** and seamlessly integrate this information into the **Previous Reasoning Steps** to continue reasoning for the original question.
+
+    **Guidelines:**
+
+    1. **Analyze the Searched Web Pages:**
+    - Carefully review the content of each searched web page.
+    - Identify factual information that is relevant to the **Current Search Query** and can aid in the reasoning process for the original question.
+
+    2. **Extract Relevant Information:**
+    - Select the information from the Searched Web Pages that directly contributes to advancing the **Previous Reasoning Steps**.
+    - Ensure that the extracted information is accurate and relevant.
+
+    3. **Output Format:**
+    - **If the web pages provide helpful information for current search query:** Present the information beginning with `**Final Information**` as shown below.
+    - The language of query **MUST BE** as the same as 'Search Query' or 'Web Pages'.\n"
+    **Final Information**
+
+    [Helpful information]
+
+    - **If the web pages do not provide any helpful information for current search query:** Output the following text.
+
+    **Final Information**
+
+    No helpful information found.
+
+    **Inputs:**
+    - **Previous Reasoning Steps:**  
+    {prev_reasoning}
+
+    - **Current Search Query:**  
+    {search_query}
+
+    - **Searched Web Pages:**  
+    {document}
+
+    """

api/apps/__init__.py

@@ -119,8 +119,9 @@ def register_page(page_path):
     sys.modules[module_name] = page
     spec.loader.exec_module(page)
     page_name = getattr(page, "page_name", page_name)
+    sdk_path = "\\sdk\\" if sys.platform.startswith("win") else "/sdk/"
     url_prefix = (
-        f"/api/{API_VERSION}" if "/sdk/" in path else f"/{API_VERSION}/{page_name}"
+        f"/api/{API_VERSION}" if sdk_path in path else f"/{API_VERSION}/{page_name}"
     )
 
     app.register_blueprint(page.manager, url_prefix=url_prefix)

api/apps/api_app.py

@@ -25,7 +25,7 @@ from api.db import FileType, LLMType, ParserType, FileSource
 from api.db.db_models import APIToken, Task, File
 from api.db.services import duplicate_name
 from api.db.services.api_service import APITokenService, API4ConversationService
-from api.db.services.dialog_service import DialogService, chat, keyword_extraction, label_question
+from api.db.services.dialog_service import DialogService, chat
 from api.db.services.document_service import DocumentService, doc_upload_and_parse
 from api.db.services.file2document_service import File2DocumentService
 from api.db.services.file_service import FileService
@@ -38,6 +38,8 @@ from api.utils.api_utils import server_error_response, get_data_error_result, ge
     generate_confirmation_token
 
 from api.utils.file_utils import filename_type, thumbnail
+from rag.app.tag import label_question
+from rag.prompts import keyword_extraction
 from rag.utils.storage_factory import STORAGE_IMPL
 
 from api.db.services.canvas_service import UserCanvasService

api/apps/chunk_app.py

@@ -19,9 +19,10 @@ import json
 from flask import request
 from flask_login import login_required, current_user
 
-from api.db.services.dialog_service import keyword_extraction, label_question
 from rag.app.qa import rmPrefix, beAdoc
+from rag.app.tag import label_question
 from rag.nlp import search, rag_tokenizer
+from rag.prompts import keyword_extraction
 from rag.settings import PAGERANK_FLD
 from rag.utils import rmSpace
 from api.db import LLMType, ParserType
@@ -93,12 +94,14 @@ def get():
         tenants = UserTenantService.query(user_id=current_user.id)
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
-        tenant_id = tenants[0].tenant_id
-
-        kb_ids = KnowledgebaseService.get_kb_ids(tenant_id)
-        chunk = settings.docStoreConn.get(chunk_id, search.index_name(tenant_id), kb_ids)
+        for tenant in tenants:
+            kb_ids = KnowledgebaseService.get_kb_ids(tenant.tenant_id)
+            chunk = settings.docStoreConn.get(chunk_id, search.index_name(tenant.tenant_id), kb_ids)
+            if chunk:
+                break
         if chunk is None:
             return server_error_response(Exception("Chunk not found"))
+
         k = []
         for n in chunk.keys():
             if re.search(r"(_vec$|_sm_|_tks|_ltks)", n):

api/apps/conversation_app.py

@@ -17,6 +17,7 @@ import json
 import re
 import traceback
 from copy import deepcopy
+import trio
 from api.db.db_models import APIToken
 
 from api.db.services.conversation_service import ConversationService, structure_answer
@@ -25,13 +26,14 @@ from flask import request, Response
 from flask_login import login_required, current_user
 
 from api.db import LLMType
-from api.db.services.dialog_service import DialogService, chat, ask, label_question
+from api.db.services.dialog_service import DialogService, chat, ask
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle, TenantService
 from api import settings
 from api.utils.api_utils import get_json_result
 from api.utils.api_utils import server_error_response, get_data_error_result, validate_request
 from graphrag.general.mind_map_extractor import MindMapExtractor
+from rag.app.tag import label_question
 
 
 @manager.route('/set', methods=['POST'])  # noqa: F821
@@ -385,7 +387,8 @@ def mindmap():
                                            rank_feature=label_question(question, [kb])
                                            )
     mindmap = MindMapExtractor(chat_mdl)
-    mind_map = mindmap([c["content_with_weight"] for c in ranks["chunks"]]).output
+    mind_map = trio.run(mindmap, [c["content_with_weight"] for c in ranks["chunks"]])
+    mind_map = mind_map.output
     if "error" in mind_map:
         return server_error_response(Exception(mind_map["error"]))
     return get_json_result(data=mind_map)

api/apps/dialog_app.py

@@ -18,6 +18,7 @@ from flask import request
 from flask_login import login_required, current_user
 from api.db.services.dialog_service import DialogService
 from api.db import StatusEnum
+from api.db.services.llm_service import TenantLLMService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.user_service import TenantService, UserTenantService
 from api import settings
@@ -57,11 +58,6 @@ def set_dialog():
 
     if not prompt_config["system"]:
         prompt_config["system"] = default_prompt["system"]
-    # if len(prompt_config["parameters"]) < 1:
-    #     prompt_config["parameters"] = default_prompt["parameters"]
-    # for p in prompt_config["parameters"]:
-    #     if p["key"] == "knowledge":break
-    # else: prompt_config["parameters"].append(default_prompt["parameters"][0])
 
     for p in prompt_config["parameters"]:
         if p["optional"]:
@@ -74,22 +70,19 @@ def set_dialog():
         e, tenant = TenantService.get_by_id(current_user.id)
         if not e:
             return get_data_error_result(message="Tenant not found!")
-        kbs = KnowledgebaseService.get_by_ids(req.get("kb_ids"))
-        embd_count = len(set([kb.embd_id for kb in kbs]))
-        if embd_count != 1:
+        kbs = KnowledgebaseService.get_by_ids(req.get("kb_ids", []))
+        embd_ids = [TenantLLMService.split_model_name_and_factory(kb.embd_id)[0] for kb in kbs]  # remove vendor suffix for comparison
+        embd_count = len(set(embd_ids))
+        if embd_count > 1:
             return get_data_error_result(message=f'Datasets use different embedding models: {[kb.embd_id for kb in kbs]}"')
 
         llm_id = req.get("llm_id", tenant.llm_id)
         if not dialog_id:
-            if not req.get("kb_ids"):
-                return get_data_error_result(
-                    message="Fail! Please select knowledgebase!")
-
             dia = {
                 "id": get_uuid(),
                 "tenant_id": current_user.id,
                 "name": name,
-                "kb_ids": req["kb_ids"],
+                "kb_ids": req.get("kb_ids", []),
                 "description": description,
                 "llm_id": llm_id,
                 "llm_setting": llm_setting,

api/apps/document_app.py

@@ -71,11 +71,13 @@ def upload():
     if not e:
         raise LookupError("Can't find this knowledgebase!")
 
-    err, _ = FileService.upload_document(kb, file_objs, current_user.id)
+    err, files = FileService.upload_document(kb, file_objs, current_user.id)
+    files = [f[0] for f in files] # remove the blob
+    
     if err:
         return get_json_result(
-            data=False, message="\n".join(err), code=settings.RetCode.SERVER_ERROR)
-    return get_json_result(data=True)
+            data=files, message="\n".join(err), code=settings.RetCode.SERVER_ERROR)
+    return get_json_result(data=files)
 
 
 @manager.route('/web_crawl', methods=['POST'])  # noqa: F821
@@ -345,7 +347,7 @@ def rm():
 @manager.route('/run', methods=['POST'])  # noqa: F821
 @login_required
 @validate_request("doc_ids", "run")
-def run():
+def run(): 
     req = request.json
     for doc_id in req["doc_ids"]:
         if not DocumentService.accessible(doc_id, current_user.id):

api/apps/kb_app.py

@@ -14,7 +14,6 @@
 #  limitations under the License.
 #
 import json
-import logging
 import os
 
 from flask import request
@@ -300,11 +299,12 @@ def knowledge_graph(kb_id):
         "kb_id": [kb_id],
         "knowledge_graph_kwd": ["graph"]
     }
+
     obj = {"graph": {}, "mind_map": {}}
-    try:
-        sres = settings.retrievaler.search(req, search.index_name(kb.tenant_id), [kb_id])
-    except Exception as e:
-        logging.exception(e)
+    if not settings.docStoreConn.indexExist(search.index_name(kb.tenant_id), kb_id):
+        return get_json_result(data=obj)
+    sres = settings.retrievaler.search(req, search.index_name(kb.tenant_id), [kb_id])
+    if not len(sres.ids):
         return get_json_result(data=obj)
 
     for id in sres.ids[:1]:
@@ -318,6 +318,8 @@ def knowledge_graph(kb_id):
 
     if "nodes" in obj["graph"]:
         obj["graph"]["nodes"] = sorted(obj["graph"]["nodes"], key=lambda x: x.get("pagerank", 0), reverse=True)[:256]
-    if "edges" in obj["graph"]:
-        obj["graph"]["edges"] = sorted(obj["graph"]["edges"], key=lambda x: x.get("weight", 0), reverse=True)[:128]
+        if "edges" in obj["graph"]:
+            node_id_set = { o["id"] for o in obj["graph"]["nodes"] }
+            filtered_edges = [o for o in obj["graph"]["edges"] if o["source"] != o["target"] and o["source"] in node_id_set and o["target"] in node_id_set]
+            obj["graph"]["edges"] = sorted(filtered_edges, key=lambda x: x.get("weight", 0), reverse=True)[:128]
     return get_json_result(data=obj)

api/apps/llm_app.py

@@ -135,6 +135,8 @@ def set_api_key():
 def add_llm():
     req = request.json
     factory = req["llm_factory"]
+    api_key = req.get("api_key", "x")
+    llm_name = req["llm_name"]
 
     def apikey_json(keys):
         nonlocal req
@@ -143,7 +145,6 @@ def add_llm():
     if factory == "VolcEngine":
         # For VolcEngine, due to its special authentication method
         # Assemble ark_api_key endpoint_id into api_key
-        llm_name = req["llm_name"]
         api_key = apikey_json(["ark_api_key", "endpoint_id"])
 
     elif factory == "Tencent Hunyuan":
@@ -152,52 +153,43 @@ def add_llm():
 
     elif factory == "Tencent Cloud":
         req["api_key"] = apikey_json(["tencent_cloud_sid", "tencent_cloud_sk"])
+        return set_api_key()
 
     elif factory == "Bedrock":
         # For Bedrock, due to its special authentication method
         # Assemble bedrock_ak, bedrock_sk, bedrock_region
-        llm_name = req["llm_name"]
         api_key = apikey_json(["bedrock_ak", "bedrock_sk", "bedrock_region"])
 
     elif factory == "LocalAI":
-        llm_name = req["llm_name"] + "___LocalAI"
-        api_key = "xxxxxxxxxxxxxxx"
+        llm_name += "___LocalAI"
 
     elif factory == "HuggingFace":
-        llm_name = req["llm_name"] + "___HuggingFace"
-        api_key = "xxxxxxxxxxxxxxx"
+        llm_name += "___HuggingFace"
 
     elif factory == "OpenAI-API-Compatible":
-        llm_name = req["llm_name"] + "___OpenAI-API"
-        api_key = req.get("api_key", "xxxxxxxxxxxxxxx")
+        llm_name += "___OpenAI-API"
+
+    elif factory == "VLLM":
+        llm_name += "___VLLM"
 
     elif factory == "XunFei Spark":
-        llm_name = req["llm_name"]
         if req["model_type"] == "chat":
-            api_key = req.get("spark_api_password", "xxxxxxxxxxxxxxx")
+            api_key = req.get("spark_api_password", "")
         elif req["model_type"] == "tts":
             api_key = apikey_json(["spark_app_id", "spark_api_secret", "spark_api_key"])
 
     elif factory == "BaiduYiyan":
-        llm_name = req["llm_name"]
         api_key = apikey_json(["yiyan_ak", "yiyan_sk"])
 
     elif factory == "Fish Audio":
-        llm_name = req["llm_name"]
         api_key = apikey_json(["fish_audio_ak", "fish_audio_refid"])
 
     elif factory == "Google Cloud":
-        llm_name = req["llm_name"]
         api_key = apikey_json(["google_project_id", "google_region", "google_service_account_key"])
 
     elif factory == "Azure-OpenAI":
-        llm_name = req["llm_name"]
         api_key = apikey_json(["api_key", "api_version"])
 
-    else:
-        llm_name = req["llm_name"]
-        api_key = req.get("api_key", "xxxxxxxxxxxxxxx")
-
     llm = {
         "tenant_id": current_user.id,
         "llm_factory": factory,
@@ -209,66 +201,69 @@ def add_llm():
     }
 
     msg = ""
+    mdl_nm = llm["llm_name"].split("___")[0]
     if llm["model_type"] == LLMType.EMBEDDING.value:
         mdl = EmbeddingModel[factory](
             key=llm['api_key'],
-            model_name=llm["llm_name"],
+            model_name=mdl_nm,
             base_url=llm["api_base"])
         try:
             arr, tc = mdl.encode(["Test if the api key is available"])
             if len(arr[0]) == 0:
                 raise Exception("Fail")
         except Exception as e:
-            msg += f"\nFail to access embedding model({llm['llm_name']})." + str(e)
+            msg += f"\nFail to access embedding model({mdl_nm})." + str(e)
     elif llm["model_type"] == LLMType.CHAT.value:
         mdl = ChatModel[factory](
             key=llm['api_key'],
-            model_name=llm["llm_name"],
+            model_name=mdl_nm,
             base_url=llm["api_base"]
         )
         try:
             m, tc = mdl.chat(None, [{"role": "user", "content": "Hello! How are you doing!"}], {
                 "temperature": 0.9})
-            if not tc:
+            if not tc and m.find("**ERROR**:") >= 0:
                 raise Exception(m)
         except Exception as e:
-            msg += f"\nFail to access model({llm['llm_name']})." + str(
+            msg += f"\nFail to access model({mdl_nm})." + str(
                 e)
     elif llm["model_type"] == LLMType.RERANK:
-        mdl = RerankModel[factory](
-            key=llm["api_key"],
-            model_name=llm["llm_name"],
-            base_url=llm["api_base"]
-        )
         try:
+            mdl = RerankModel[factory](
+                key=llm["api_key"],
+                model_name=mdl_nm,
+                base_url=llm["api_base"]
+            )
             arr, tc = mdl.similarity("Hello~ Ragflower!", ["Hi, there!", "Ohh, my friend!"])
             if len(arr) == 0:
                 raise Exception("Not known.")
+        except KeyError:
+            msg += f"{factory} dose not support this model({mdl_nm})"
         except Exception as e:
-            msg += f"\nFail to access model({llm['llm_name']})." + str(
+            msg += f"\nFail to access model({mdl_nm})." + str(
                 e)
     elif llm["model_type"] == LLMType.IMAGE2TEXT.value:
         mdl = CvModel[factory](
             key=llm["api_key"],
-            model_name=llm["llm_name"],
+            model_name=mdl_nm,
             base_url=llm["api_base"]
         )
         try:
             with open(os.path.join(get_project_base_directory(), "web/src/assets/yay.jpg"), "rb") as f:
                 m, tc = mdl.describe(f.read())
-                if not tc:
+                if not m and not tc:
                     raise Exception(m)
         except Exception as e:
-            msg += f"\nFail to access model({llm['llm_name']})." + str(e)
+            msg += f"\nFail to access model({mdl_nm})." + str(e)
     elif llm["model_type"] == LLMType.TTS:
         mdl = TTSModel[factory](
-            key=llm["api_key"], model_name=llm["llm_name"], base_url=llm["api_base"]
+            key=llm["api_key"], model_name=mdl_nm, base_url=llm["api_base"]
         )
         try:
             for resp in mdl.tts("Hello~ Ragflower!"):
                 pass
         except RuntimeError as e:
-            msg += f"\nFail to access model({llm['llm_name']})." + str(e)
+            msg += f"\nFail to access model({mdl_nm})." + str(e)
     else:
         # TODO: check other type of models
         pass
@@ -343,8 +338,6 @@ def list_app():
 
         llm_set = set([m["llm_name"] + "@" + m["fid"] for m in llms])
         for o in objs:
-            if not o.api_key:
-                continue
             if o.llm_name + "@" + o.llm_factory in llm_set:
                 continue
             llms.append({"llm_name": o.llm_name, "model_type": o.model_type, "fid": o.llm_factory, "available": True})
@@ -359,4 +352,4 @@ def list_app():
 
         return get_json_result(data=res)
     except Exception as e:
-        return server_error_response(e)
+        return server_error_response(e)

api/apps/sdk/chat.py

@@ -13,6 +13,8 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
+import logging
+
 from flask import request
 from api import settings
 from api.db import StatusEnum
@@ -29,9 +31,7 @@ from api.utils.api_utils import get_result
 @token_required
 def create(tenant_id):
     req = request.json
-    ids = req.get("dataset_ids")
-    if not ids:
-        return get_error_data_result(message="`dataset_ids` is required")
+    ids = [i for i in req.get("dataset_ids", []) if i] 
     for kb_id in ids:
         kbs = KnowledgebaseService.accessible(kb_id=kb_id, user_id=tenant_id)
         if not kbs:
@@ -40,9 +40,16 @@ def create(tenant_id):
         kb = kbs[0]
         if kb.chunk_num == 0:
             return get_error_data_result(f"The dataset {kb_id} doesn't own parsed file")
-    kbs = KnowledgebaseService.get_by_ids(ids)
-    embd_count = list(set([kb.embd_id for kb in kbs]))
-    if len(embd_count) != 1:
+        
+        # Check if all documents in the knowledge base have been parsed
+        is_done, error_msg = KnowledgebaseService.is_parsed_done(kb_id)
+        if not is_done:
+            return get_error_data_result(error_msg)
+    
+    kbs = KnowledgebaseService.get_by_ids(ids) if ids else []
+    embd_ids = [TenantLLMService.split_model_name_and_factory(kb.embd_id)[0] for kb in kbs]  # remove vendor suffix for comparison
+    embd_count = list(set(embd_ids))
+    if len(embd_count) > 1:
         return get_result(message='Datasets use different embedding models."',
                           code=settings.RetCode.AUTHENTICATION_ERROR)
     req["kb_ids"] = ids
@@ -175,8 +182,15 @@ def update(tenant_id, chat_id):
                 kb = kbs[0]
                 if kb.chunk_num == 0:
                     return get_error_data_result(f"The dataset {kb_id} doesn't own parsed file")
+                
+                # Check if all documents in the knowledge base have been parsed
+                is_done, error_msg = KnowledgebaseService.is_parsed_done(kb_id)
+                if not is_done:
+                    return get_error_data_result(error_msg)
+                
             kbs = KnowledgebaseService.get_by_ids(ids)
-            embd_count = list(set([kb.embd_id for kb in kbs]))
+            embd_ids = [TenantLLMService.split_model_name_and_factory(kb.embd_id)[0] for kb in kbs]  # remove vendor suffix for comparison
+            embd_count = list(set(embd_ids))
             if len(embd_count) != 1:
                 return get_result(
                     message='Datasets use different embedding models."',
@@ -316,7 +330,8 @@ def list_chat(tenant_id):
         for kb_id in res["kb_ids"]:
             kb = KnowledgebaseService.query(id=kb_id)
             if not kb:
-                return get_error_data_result(message=f"Don't exist the kb {kb_id}")
+                logging.warning(f"The kb {kb_id} does not exist.")
+                continue
             kb_list.append(kb[0].to_json())
         del res["kb_ids"]
         res["datasets"] = kb_list

api/apps/sdk/dataset.py

@@ -30,7 +30,7 @@ from api.utils.api_utils import (
     token_required,
     get_error_data_result,
     valid,
-    get_parser_config,
+    get_parser_config, valid_parser_config,
 )
 
 
@@ -66,10 +66,6 @@ def create(tenant_id):
               type: string
               enum: ['me', 'team']
               description: Dataset permission.
-            language:
-              type: string
-              enum: ['Chinese', 'English']
-              description: Language of the dataset.
             chunk_method:
               type: string
               enum: ["naive", "manual", "qa", "table", "paper", "book", "laws",
@@ -91,11 +87,10 @@ def create(tenant_id):
     req = request.json
     e, t = TenantService.get_by_id(tenant_id)
     permission = req.get("permission")
-    language = req.get("language")
     chunk_method = req.get("chunk_method")
     parser_config = req.get("parser_config")
+    valid_parser_config(parser_config)
     valid_permission = ["me", "team"]
-    valid_language = ["Chinese", "English"]
     valid_chunk_method = [
         "naive",
         "manual",
@@ -114,8 +109,6 @@ def create(tenant_id):
     check_validation = valid(
         permission,
         valid_permission,
-        language,
-        valid_language,
         chunk_method,
         valid_chunk_method,
     )
@@ -134,13 +127,18 @@ def create(tenant_id):
     req["name"] = req["name"].strip()
     if req["name"] == "":
         return get_error_data_result(message="`name` is not empty string!")
+    if len(req["name"]) >= 128:
+        return get_error_data_result(
+            message="Dataset name should not be longer than 128 characters."
+        )
     if KnowledgebaseService.query(
         name=req["name"], tenant_id=tenant_id, status=StatusEnum.VALID.value
     ):
         return get_error_data_result(
             message="Duplicated dataset name in creating dataset."
         )
-    req["tenant_id"] = req["created_by"] = tenant_id
+    req["tenant_id"] = tenant_id
+    req["created_by"] = tenant_id
     if not req.get("embedding_model"):
         req["embedding_model"] = t.embd_id
     else:
@@ -182,6 +180,10 @@ def create(tenant_id):
         if old_key in req
     }
     req.update(mapped_keys)
+    flds = list(req.keys())
+    for f in flds:
+        if req[f] == "" and f in ["permission", "parser_id", "chunk_method"]:
+            del req[f]
     if not KnowledgebaseService.save(**req):
         return get_error_data_result(message="Create dataset error.(Database error)")
     renamed_data = {}
@@ -226,6 +228,8 @@ def delete(tenant_id):
         schema:
           type: object
     """
+    errors = []
+    success_count = 0
     req = request.json
     if not req:
         ids = None
@@ -241,12 +245,12 @@ def delete(tenant_id):
     for id in id_list:
         kbs = KnowledgebaseService.query(id=id, tenant_id=tenant_id)
         if not kbs:
-            return get_error_data_result(message=f"You don't own the dataset {id}")
+            errors.append(f"You don't own the dataset {id}")
+            continue
         for doc in DocumentService.query(kb_id=id):
             if not DocumentService.remove_document(doc, tenant_id):
-                return get_error_data_result(
-                    message="Remove document error.(Database error)"
-                )
+                errors.append(f"Remove document error for dataset {id}")
+                continue
             f2d = File2DocumentService.get_by_document_id(doc.id)
             FileService.filter_delete(
                 [
@@ -258,11 +262,21 @@ def delete(tenant_id):
         FileService.filter_delete(
             [File.source_type == FileSource.KNOWLEDGEBASE, File.type == "folder", File.name == kbs[0].name])
         if not KnowledgebaseService.delete_by_id(id):
-            return get_error_data_result(message="Delete dataset error.(Database error)")
+            errors.append(f"Delete dataset error for {id}")
+            continue
+        success_count += 1
+    if errors:
+        if success_count > 0:
+            return get_result(
+                data={"success_count": success_count, "errors": errors},
+                message=f"Partially deleted {success_count} datasets with {len(errors)} errors"
+            )
+        else:
+            return get_error_data_result(message="; ".join(errors))
     return get_result(code=settings.RetCode.SUCCESS)
 
 
-@manager.route("/datasets/<dataset_id>", methods=["PUT"])  # noqa: F821
+@manager.route("/datasets/<dataset_id>", methods=["PUT"])  # noqa: F821  
 @token_required
 def update(tenant_id, dataset_id):
     """
@@ -297,10 +311,6 @@ def update(tenant_id, dataset_id):
               type: string
               enum: ['me', 'team']
               description: Updated permission.
-            language:
-              type: string
-              enum: ['Chinese', 'English']
-              description: Updated language.
             chunk_method:
               type: string
               enum: ["naive", "manual", "qa", "table", "paper", "book", "laws",
@@ -320,15 +330,14 @@ def update(tenant_id, dataset_id):
         return get_error_data_result(message="You don't own the dataset")
     req = request.json
     e, t = TenantService.get_by_id(tenant_id)
-    invalid_keys = {"id", "embd_id", "chunk_num", "doc_num", "parser_id"}
+    invalid_keys = {"id", "embd_id", "chunk_num", "doc_num", "parser_id", "create_date", "create_time", "created_by", "status","token_num","update_date","update_time"}
     if any(key in req for key in invalid_keys):
         return get_error_data_result(message="The input parameters are invalid.")
     permission = req.get("permission")
-    language = req.get("language")
     chunk_method = req.get("chunk_method")
     parser_config = req.get("parser_config")
+    valid_parser_config(parser_config)
     valid_permission = ["me", "team"]
-    valid_language = ["Chinese", "English"]
     valid_chunk_method = [
         "naive",
         "manual",
@@ -347,8 +356,6 @@ def update(tenant_id, dataset_id):
     check_validation = valid(
         permission,
         valid_permission,
-        language,
-        valid_language,
         chunk_method,
         valid_chunk_method,
     )
@@ -370,7 +377,7 @@ def update(tenant_id, dataset_id):
         if req["document_count"] != kb.doc_num:
             return get_error_data_result(message="Can't change `document_count`.")
         req.pop("document_count")
-    if "chunk_method" in req:
+    if req.get("chunk_method"):
         if kb.chunk_num != 0 and req["chunk_method"] != kb.parser_id:
             return get_error_data_result(
                 message="If `chunk_count` is not 0, `chunk_method` is not changeable."
@@ -416,6 +423,10 @@ def update(tenant_id, dataset_id):
         req["embd_id"] = req.pop("embedding_model")
     if "name" in req:
         req["name"] = req["name"].strip()
+        if len(req["name"]) >= 128:
+            return get_error_data_result(
+                message="Dataset name should not be longer than 128 characters."
+            )
         if (
             req["name"].lower() != kb.name.lower()
             and len(
@@ -428,6 +439,10 @@ def update(tenant_id, dataset_id):
             return get_error_data_result(
                 message="Duplicated dataset name in updating dataset."
             )
+    flds = list(req.keys())
+    for f in flds:
+        if req[f] == "" and f in ["permission", "parser_id", "chunk_method"]:
+            del req[f]
     if not KnowledgebaseService.update_by_id(kb.id, req):
         return get_error_data_result(message="Update dataset error.(Database error)")
     return get_result(code=settings.RetCode.SUCCESS)
@@ -435,7 +450,7 @@ def update(tenant_id, dataset_id):
 
 @manager.route("/datasets", methods=["GET"])  # noqa: F821
 @token_required
-def list(tenant_id):
+def list_datasets(tenant_id):
     """
     List datasets.
     ---
@@ -504,7 +519,9 @@ def list(tenant_id):
     page_number = int(request.args.get("page", 1))
     items_per_page = int(request.args.get("page_size", 30))
     orderby = request.args.get("orderby", "create_time")
-    if request.args.get("desc") == "False" or request.args.get("desc") == "false":
+    if request.args.get("desc", "false").lower() not in ["true", "false"]:
+        return get_error_data_result("desc should be true or false")
+    if request.args.get("desc", "true").lower() == "false":
         desc = False
     else:
         desc = True

api/apps/sdk/dify_retrieval.py

@@ -16,11 +16,11 @@
 from flask import request, jsonify
 
 from api.db import LLMType
-from api.db.services.dialog_service import label_question
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from api.utils.api_utils import validate_request, build_error_result, apikey_required
+from rag.app.tag import label_question
 
 
 @manager.route('/dify/retrieval', methods=['POST'])  # noqa: F821

api/apps/sdk/doc.py

@@ -16,7 +16,6 @@
 import pathlib
 import datetime
 
-from api.db.services.dialog_service import keyword_extraction, label_question
 from rag.app.qa import rmPrefix, beAdoc
 from rag.nlp import rag_tokenizer
 from api.db import LLMType, ParserType
@@ -39,6 +38,8 @@ from api.db.services.file_service import FileService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.utils.api_utils import construct_json_result, get_parser_config
 from rag.nlp import search
+from rag.prompts import keyword_extraction
+from rag.app.tag import label_question
 from rag.utils import rmSpace
 from rag.utils.storage_factory import STORAGE_IMPL
 
@@ -239,6 +240,11 @@ def update_doc(tenant_id, dataset_id, document_id):
         if req["progress"] != doc.progress:
             return get_error_data_result(message="Can't change `progress`.")
 
+    if "meta_fields" in req:
+        if not isinstance(req["meta_fields"], dict):
+            return get_error_data_result(message="meta_fields must be a dictionary")
+        DocumentService.update_meta_fields(document_id, req["meta_fields"])
+
     if "name" in req and req["name"] != doc.name:
         if (
                 pathlib.Path(req["name"].lower()).suffix
@@ -260,6 +266,7 @@ def update_doc(tenant_id, dataset_id, document_id):
         if informs:
             e, file = FileService.get_by_id(informs[0].file_id)
             FileService.update_by_id(file.id, {"name": req["name"]})
+
     if "parser_config" in req:
         DocumentService.update_parser_config(doc.id, req["parser_config"])
     if "chunk_method" in req:
@@ -472,10 +479,12 @@ def list_docs(dataset_id, tenant_id):
         return get_error_data_result(message=f"You don't own the dataset {dataset_id}. ")
     id = request.args.get("id")
     name = request.args.get("name")
-    if not DocumentService.query(id=id, kb_id=dataset_id):
+
+    if id and not DocumentService.query(id=id, kb_id=dataset_id):
         return get_error_data_result(message=f"You don't own the document {id}.")
-    if not DocumentService.query(name=name, kb_id=dataset_id):
+    if name and not DocumentService.query(name=name, kb_id=dataset_id):
         return get_error_data_result(message=f"You don't own the document {name}.")
+
     page = int(request.args.get("page", 1))
     keywords = request.args.get("keywords", "")
     page_size = int(request.args.get("page_size", 30))
@@ -729,7 +738,7 @@ def stop_parsing(tenant_id, dataset_id):
             )
         info = {"run": "2", "progress": 0, "chunk_num": 0}
         DocumentService.update_by_id(id, info)
-        settings.docStoreConn.delete({"doc_id": doc.id}, search.index_name(tenant_id), dataset_id)
+        settings.docStoreConn.delete({"doc_id": doc[0].id}, search.index_name(tenant_id), dataset_id)
     return get_result()
 
 
@@ -1301,7 +1310,7 @@ def retrieval_test(tenant_id):
         if not KnowledgebaseService.accessible(kb_id=id, user_id=tenant_id):
             return get_error_data_result(f"You don't own the dataset {id}.")
     kbs = KnowledgebaseService.get_by_ids(kb_ids)
-    embd_nms = list(set([kb.embd_id for kb in kbs]))
+    embd_nms = list(set([TenantLLMService.split_model_name_and_factory(kb.embd_id)[0] for kb in kbs]))  # remove vendor suffix for comparison
     if len(embd_nms) != 1:
         return get_result(
             message='Datasets use different embedding models."',

api/apps/sdk/session.py

@@ -15,13 +15,13 @@
 #
 import re
 import json
-from api.db import LLMType
-from flask import request, Response
+import time
 
+from api.db import LLMType
 from api.db.services.conversation_service import ConversationService, iframe_completion
 from api.db.services.conversation_service import completion as rag_completion
 from api.db.services.canvas_service import completion as agent_completion
-from api.db.services.dialog_service import ask
+from api.db.services.dialog_service import ask, chat
 from agent.canvas import Canvas
 from api.db import StatusEnum
 from api.db.db_models import APIToken
@@ -30,11 +30,12 @@ from api.db.services.canvas_service import UserCanvasService
 from api.db.services.dialog_service import DialogService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.utils import get_uuid
-from api.utils.api_utils import get_error_data_result
+from api.utils.api_utils import get_error_data_result, validate_request
 from api.utils.api_utils import get_result, token_required
 from api.db.services.llm_service import LLMBundle
+from api.db.services.file_service import FileService
 
-
+from flask import jsonify, request, Response
 
 @manager.route('/chats/<chat_id>/sessions', methods=['POST'])  # noqa: F821
 @token_required
@@ -68,6 +69,11 @@ def create(tenant_id, chat_id):
 @token_required
 def create_agent_session(tenant_id, agent_id):
     req = request.json
+    if not request.is_json:
+        req = request.form
+    files = request.files
+    user_id = request.args.get('user_id', '')
+
     e, cvs = UserCanvasService.get_by_id(agent_id)
     if not e:
         return get_error_data_result("Agent not found.")
@@ -84,15 +90,33 @@ def create_agent_session(tenant_id, agent_id):
     if query:
         for ele in query:
             if not ele["optional"]:
-                if not req.get(ele["key"]):
-                    return get_error_data_result(f"`{ele['key']}` is required")
-                ele["value"] = req[ele["key"]]
-            if ele["optional"]:
-                if req.get(ele["key"]):
-                    ele["value"] = req[ele['key']]
+                if ele["type"] == "file":
+                    if files is None or not files.get(ele["key"]):
+                        return get_error_data_result(f"`{ele['key']}` with type `{ele['type']}` is required")
+                    upload_file = files.get(ele["key"])
+                    file_content = FileService.parse_docs([upload_file], user_id)
+                    file_name = upload_file.filename
+                    ele["value"] = file_name + "\n" + file_content
                 else:
-                    if "value" in ele:
-                        ele.pop("value")
+                    if req is None or not req.get(ele["key"]):
+                        return get_error_data_result(f"`{ele['key']}` with type `{ele['type']}` is required")
+                    ele["value"] = req[ele["key"]]
+            else:
+                if ele["type"] == "file":
+                    if files is not None and files.get(ele["key"]):
+                        upload_file = files.get(ele["key"])
+                        file_content = FileService.parse_docs([upload_file], user_id)
+                        file_name = upload_file.filename
+                        ele["value"] = file_name + "\n" + file_content
+                    else:
+                        if "value" in ele:
+                            ele.pop("value")
+                else:
+                    if req is not None and req.get(ele["key"]):
+                        ele["value"] = req[ele['key']]
+                    else:
+                        if "value" in ele:
+                            ele.pop("value")
     else:
         for ans in canvas.run(stream=False):
             pass
@@ -100,7 +124,7 @@ def create_agent_session(tenant_id, agent_id):
     conv = {
         "id": get_uuid(),
         "dialog_id": cvs.id,
-        "user_id": req.get("user_id", "") if isinstance(req, dict) else "",
+        "user_id": user_id,
         "message": [{"role": "assistant", "content": canvas.get_prologue()}],
         "source": "agent",
         "dsl": cvs.dsl
@@ -136,8 +160,10 @@ def update(tenant_id, chat_id, session_id):
 @token_required
 def chat_completion(tenant_id, chat_id):
     req = request.json
-    if not req or not req.get("session_id"):
+    if not req:
         req = {"question": ""}
+    if not req.get("session_id"):
+        req["question"]=""
     if not DialogService.query(tenant_id=tenant_id, id=chat_id, status=StatusEnum.VALID.value):
         return get_error_data_result(f"You don't own the chat {chat_id}")
     if req.get("session_id"):
@@ -159,6 +185,175 @@ def chat_completion(tenant_id, chat_id):
         return get_result(data=answer)
 
 
+@manager.route('chats_openai/<chat_id>/chat/completions', methods=['POST'])  # noqa: F821
+@validate_request("model", "messages")  # noqa: F821
+@token_required
+def chat_completion_openai_like(tenant_id, chat_id):
+    """
+    OpenAI-like chat completion API that simulates the behavior of OpenAI's completions endpoint.
+    
+    This function allows users to interact with a model and receive responses based on a series of historical messages.
+    If `stream` is set to True (by default), the response will be streamed in chunks, mimicking the OpenAI-style API.
+    Set `stream` to False explicitly, the response will be returned in a single complete answer.
+    Example usage:
+
+    curl -X POST https://ragflow_address.com/api/v1/chats_openai/<chat_id>/chat/completions \
+        -H "Content-Type: application/json" \
+        -H "Authorization: Bearer $RAGFLOW_API_KEY" \
+        -d '{
+            "model": "model",
+            "messages": [{"role": "user", "content": "Say this is a test!"}],
+            "stream": true
+        }'
+
+    Alternatively, you can use Python's `OpenAI` client:
+
+    from openai import OpenAI
+
+    model = "model"
+    client = OpenAI(api_key="ragflow-api-key", base_url=f"http://ragflow_address/api/v1/chats_openai/<chat_id>")
+    
+    completion = client.chat.completions.create(
+        model=model,
+        messages=[
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "Who are you?"},
+            {"role": "assistant", "content": "I am an AI assistant named..."},
+            {"role": "user", "content": "Can you tell me how to install neovim"},
+        ],
+        stream=True
+    )
+    
+    stream = True
+    if stream:
+        for chunk in completion:
+            print(chunk)
+    else:
+        print(completion.choices[0].message.content)
+    """
+    req = request.json
+
+    messages = req.get("messages", [])
+    # To prevent empty [] input
+    if len(messages) < 1:
+        return get_error_data_result("You have to provide messages.")
+    if messages[-1]["role"] != "user":
+        return get_error_data_result("The last content of this conversation is not from user.")
+
+    prompt = messages[-1]["content"]
+    # Treat context tokens as reasoning tokens
+    context_token_used = sum(len(message["content"]) for message in messages)
+
+    dia = DialogService.query(tenant_id=tenant_id, id=chat_id, status=StatusEnum.VALID.value)
+    if not dia:
+        return get_error_data_result(f"You don't own the chat {chat_id}")
+    dia = dia[0]
+
+    # Filter system and non-sense assistant messages
+    msg = None
+    msg = [m for m in messages if m["role"] != "system" and (m["role"] != "assistant" or msg)]
+
+    if req.get("stream", True):
+        # The value for the usage field on all chunks except for the last one will be null.
+        # The usage field on the last chunk contains token usage statistics for the entire request.
+        # The choices field on the last chunk will always be an empty array [].
+        def streamed_response_generator(chat_id, dia, msg):
+            token_used = 0
+            should_split_index = 0
+            response = {
+                "id": f"chatcmpl-{chat_id}",
+                "choices": [
+                    {
+                        "delta": {
+                            "content": "",
+                            "role": "assistant",
+                            "function_call": None,
+                            "tool_calls": None
+                        },
+                        "finish_reason": None,
+                        "index": 0,
+                        "logprobs": None
+                    }
+                ],
+                "created": int(time.time()),
+                "model": "model",
+                "object": "chat.completion.chunk",
+                "system_fingerprint": "",
+                "usage": None
+            }
+
+            try:
+                for ans in chat(dia, msg, True):
+                    answer = ans["answer"]
+                    incremental = answer[should_split_index:]
+                    token_used += len(incremental)
+                    if incremental.endswith("</think>"):
+                        response_data_len = len(incremental.rstrip("</think>"))
+                    else:
+                        response_data_len = len(incremental)
+                    should_split_index += response_data_len
+                    response["choices"][0]["delta"]["content"] = incremental
+                    yield f"data:{json.dumps(response, ensure_ascii=False)}\n\n"
+            except Exception as e:
+                response["choices"][0]["delta"]["content"] = "**ERROR**: " + str(e)
+                yield f"data:{json.dumps(response, ensure_ascii=False)}\n\n"
+
+            # The last chunk
+            response["choices"][0]["delta"]["content"] = None
+            response["choices"][0]["finish_reason"] = "stop"
+            response["usage"] = {
+                "prompt_tokens": len(prompt),
+                "completion_tokens": token_used,
+                "total_tokens": len(prompt) + token_used
+            }
+            yield f"data:{json.dumps(response, ensure_ascii=False)}\n\n"
+            yield "data:[DONE]\n\n"
+
+
+        resp = Response(streamed_response_generator(chat_id, dia, msg), mimetype="text/event-stream")
+        resp.headers.add_header("Cache-control", "no-cache")
+        resp.headers.add_header("Connection", "keep-alive")
+        resp.headers.add_header("X-Accel-Buffering", "no")
+        resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
+        return resp
+    else:
+        answer = None
+        for ans in chat(dia, msg, False):
+            # focus answer content only
+            answer = ans
+            break
+        content = answer["answer"]
+
+        response  = {
+            "id": f"chatcmpl-{chat_id}",
+            "object": "chat.completion",
+            "created": int(time.time()),
+            "model": req.get("model", ""),
+            "usage": {
+                "prompt_tokens": len(prompt),
+                "completion_tokens": len(content),
+                "total_tokens": len(prompt) + len(content),
+                "completion_tokens_details": {
+                    "reasoning_tokens": context_token_used,
+                    "accepted_prediction_tokens": len(content),
+                    "rejected_prediction_tokens": 0 # 0 for simplicity
+                }
+            },
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "content": content
+                    },
+                    "logprobs": None,
+                    "finish_reason": "stop",
+                    "index": 0
+                }
+            ]
+        }
+        return jsonify(response)
+
+
 @manager.route('/agents/<agent_id>/completions', methods=['POST'])  # noqa: F821
 @token_required
 def agent_completions(tenant_id, agent_id):
@@ -176,6 +371,18 @@ def agent_completions(tenant_id, agent_id):
         conv = API4ConversationService.query(id=req["session_id"], dialog_id=agent_id)
         if not conv:
             return get_error_data_result(f"You don't own the session {req['session_id']}")
+        # If an update to UserCanvas is detected, update the API4Conversation.dsl
+        sync_dsl = req.get("sync_dsl", False)
+        if sync_dsl is True and cvs[0].update_time > conv[0].update_time:
+            current_dsl = conv[0].dsl
+            new_dsl = json.loads(dsl)
+            state_fields = ["history", "messages", "path", "reference"]
+            states = {field: current_dsl.get(field, []) for field in state_fields}
+            current_dsl.update(new_dsl)
+            current_dsl.update(states)
+            API4ConversationService.update_by_id(req["session_id"], {
+                "dsl": current_dsl
+            })
     else:
         req["question"] = ""
     if req.get("stream", True):
@@ -259,7 +466,10 @@ def list_agent_session(tenant_id, agent_id):
         desc = False
     else:
         desc = True
-    convs = API4ConversationService.get_list(agent_id, tenant_id, page_number, items_per_page, orderby, desc, id, user_id)
+    # dsl defaults to True in all cases except for False and false
+    include_dsl = request.args.get("dsl") != "False" and request.args.get("dsl") != "false"
+    convs = API4ConversationService.get_list(agent_id, tenant_id, page_number, items_per_page, orderby, desc, id,
+                                             user_id, include_dsl)
     if not convs:
         return get_result(data=[])
     for conv in convs:
@@ -322,6 +532,38 @@ def delete(tenant_id, chat_id):
     return get_result()
 
 
+@manager.route('/agents/<agent_id>/sessions', methods=["DELETE"])  # noqa: F821
+@token_required
+def delete_agent_session(tenant_id, agent_id):
+    req = request.json
+    cvs = UserCanvasService.query(user_id=tenant_id, id=agent_id)
+    if not cvs:
+        return get_error_data_result(f"You don't own the agent {agent_id}")
+    
+    convs = API4ConversationService.query(dialog_id=agent_id)
+    if not convs:
+        return get_error_data_result(f"Agent {agent_id} has no sessions")
+
+    if not req:
+        ids = None
+    else:
+        ids = req.get("ids")
+
+    if not ids:
+        conv_list = []
+        for conv in convs:
+            conv_list.append(conv.id)
+    else:
+        conv_list = ids
+    
+    for session_id in conv_list:
+        conv = API4ConversationService.query(id=session_id, dialog_id=agent_id)
+        if not conv:
+            return get_error_data_result(f"The agent doesn't own the session ${session_id}")
+        API4ConversationService.delete_by_id(session_id)
+    return get_result()
+    
+
 @manager.route('/sessions/ask', methods=['POST'])  # noqa: F821
 @token_required
 def ask_about(tenant_id):

api/apps/system_app.py

@@ -201,7 +201,7 @@ def new_token():
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
 
-        tenant_id = tenants[0].tenant_id
+        tenant_id = [tenant for tenant in tenants if tenant.role == 'owner'][0].tenant_id
         obj = {
             "tenant_id": tenant_id,
             "token": generate_confirmation_token(tenant_id),
@@ -256,7 +256,7 @@ def token_list():
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
 
-        tenant_id = tenants[0].tenant_id
+        tenant_id = [tenant for tenant in tenants if tenant.role == 'owner'][0].tenant_id
         objs = APITokenService.query(tenant_id=tenant_id)
         objs = [o.to_dict() for o in objs]
         for o in objs:

api/apps/user_app.py

@@ -566,7 +566,7 @@ def user_add():
     email_address = req["email"]
 
     # Validate the email address
-    if not re.match(r"^[\w\._-]+@([\w_-]+\.)+[\w-]{2,5}$", email_address):
+    if not re.match(r"^[\w\._-]+@([\w_-]+\.)+[\w-]{2,}$", email_address):
         return get_json_result(
             data=False,
             message=f"Invalid email address: {email_address}!",

api/db/db_models.py

@@ -843,8 +843,8 @@ class Task(DataBaseModel):
     id = CharField(max_length=32, primary_key=True)
     doc_id = CharField(max_length=32, null=False, index=True)
     from_page = IntegerField(default=0)
-
     to_page = IntegerField(default=100000000)
+    task_type = CharField(max_length=32, null=False, default="")
 
     begin_at = DateTimeField(null=True, index=True)
     process_duation = FloatField(default=0)
@@ -935,7 +935,7 @@ class Conversation(DataBaseModel):
 class APIToken(DataBaseModel):
     tenant_id = CharField(max_length=32, null=False, index=True)
     token = CharField(max_length=255, null=False, index=True)
-    dialog_id = CharField(max_length=32, null=False, index=True)
+    dialog_id = CharField(max_length=32, null=True, index=True)
     source = CharField(max_length=16, null=True, help_text="none|agent|dialog", index=True)
     beta = CharField(max_length=255, null=True, index=True)
 
@@ -1115,3 +1115,10 @@ def migrate_db():
             )
         except Exception:
             pass
+        try:
+            migrate(
+                migrator.add_column("task", "task_type",
+                                    CharField(max_length=32, null=False, default=""))
+            )
+        except Exception:
+            pass

api/db/init_data.py

@@ -160,7 +160,7 @@ def add_graph_templates():
     dir = os.path.join(get_project_base_directory(), "agent", "templates")
     for fnm in os.listdir(dir):
         try:
-            cnvs = json.load(open(os.path.join(dir, fnm), "r"))
+            cnvs = json.load(open(os.path.join(dir, fnm), "r",encoding="utf-8"))
             try:
                 CanvasTemplateService.save(**cnvs)
             except Exception:

api/db/services/api_service.py

@@ -43,8 +43,12 @@ class API4ConversationService(CommonService):
     @DB.connection_context()
     def get_list(cls, dialog_id, tenant_id,
                  page_number, items_per_page,
-                 orderby, desc, id, user_id=None):
-        sessions = cls.model.select().where(cls.model.dialog_id == dialog_id)
+                 orderby, desc, id, user_id=None, include_dsl=True):
+        if include_dsl:
+            sessions = cls.model.select().where(cls.model.dialog_id == dialog_id)
+        else:
+            fields = [field for field in cls.model._meta.fields.values() if field.name != 'dsl']
+            sessions = cls.model.select(*fields).where(cls.model.dialog_id == dialog_id)
         if id:
             sessions = sessions.where(cls.model.id == id)
         if user_id:

api/db/services/canvas_service.py

@@ -86,21 +86,9 @@ def completion(tenant_id, agent_id, question, session_id=None, stream=True, **kw
             "dsl": cvs.dsl
         }
         API4ConversationService.save(**conv)
-        if query:
-            yield "data:" + json.dumps({"code": 0,
-                                        "message": "",
-                                        "data": {
-                                            "session_id": session_id,
-                                            "answer": canvas.get_prologue(),
-                                            "reference": [],
-                                            "param": canvas.get_preset_param()
-                                        }
-                                        },
-                                       ensure_ascii=False) + "\n\n"
-            yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
-            return
-        else:
-            conv = API4Conversation(**conv)
+
+        
+        conv = API4Conversation(**conv)
     else:
         e, conv = API4ConversationService.get_by_id(session_id)
         assert e, "Session not found!"
@@ -130,7 +118,7 @@ def completion(tenant_id, agent_id, question, session_id=None, stream=True, **kw
                     continue
                 for k in ans.keys():
                     final_ans[k] = ans[k]
-                ans = {"answer": ans["content"], "reference": ans.get("reference", [])}
+                ans = {"answer": ans["content"], "reference": ans.get("reference", []), "param": canvas.get_preset_param()}
                 ans = structure_answer(conv, ans, message_id, session_id)
                 yield "data:" + json.dumps({"code": 0, "message": "", "data": ans},
                                            ensure_ascii=False) + "\n\n"
@@ -160,8 +148,8 @@ def completion(tenant_id, agent_id, question, session_id=None, stream=True, **kw
                 canvas.reference.append(final_ans["reference"])
             conv.dsl = json.loads(str(canvas))
 
-            result = {"answer": final_ans["content"], "reference": final_ans.get("reference", [])}
+            result = {"answer": final_ans["content"], "reference": final_ans.get("reference", []) , "param": canvas.get_preset_param()}
             result = structure_answer(conv, result, message_id, session_id)
             API4ConversationService.append_message(conv.id, conv.to_dict())
             yield result
-            break
+            break

api/db/services/conversation_service.py

@@ -23,6 +23,8 @@ from api.db.services.dialog_service import DialogService, chat
 from api.utils import get_uuid
 import json
 
+from rag.prompts import chunks_format
+
 
 class ConversationService(CommonService):
     model = Conversation
@@ -53,18 +55,7 @@ def structure_answer(conv, ans, message_id, session_id):
         reference = {}
         ans["reference"] = {}
 
-    def get_value(d, k1, k2):
-        return d.get(k1, d.get(k2))
-
-    chunk_list = [{
-        "id": get_value(chunk, "chunk_id", "id"),
-        "content": get_value(chunk, "content", "content_with_weight"),
-        "document_id": get_value(chunk, "doc_id", "document_id"),
-        "document_name": get_value(chunk, "docnm_kwd", "document_name"),
-        "dataset_id": get_value(chunk, "kb_id", "dataset_id"),
-        "image_id": get_value(chunk, "image_id", "img_id"),
-        "positions": get_value(chunk, "positions", "position_int"),
-    } for chunk in reference.get("chunks", [])]
+    chunk_list = chunks_format(reference)
 
     reference["chunks"] = chunk_list
     ans["id"] = message_id

api/db/services/dialog_service.py

@@ -15,28 +15,25 @@
 #
 import logging
 import binascii
-import os
-import json
-import json_repair
+import time
+from functools import partial
 import re
-from collections import defaultdict
 from copy import deepcopy
 from timeit import default_timer as timer
-import datetime
-from datetime import timedelta
+from agentic_reasoning import DeepResearcher
 from api.db import LLMType, ParserType, StatusEnum
 from api.db.db_models import Dialog, DB
 from api.db.services.common_service import CommonService
-from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.llm_service import LLMService, TenantLLMService, LLMBundle
+from api.db.services.llm_service import TenantLLMService, LLMBundle
 from api import settings
-from graphrag.utils import get_tags_from_cache, set_tags_to_cache
 from rag.app.resume import forbidden_select_fields4resume
+from rag.app.tag import label_question
 from rag.nlp.search import index_name
-from rag.settings import TAG_FLD
-from rag.utils import rmSpace, num_tokens_from_string, encoder
-from api.utils.file_utils import get_project_base_directory
+from rag.prompts import kb_prompt, message_fit_in, llm_id2llm_type, keyword_extraction, full_question, chunks_format, \
+    citation_prompt
+from rag.utils import rmSpace, num_tokens_from_string
+from rag.utils.tavily_conn import Tavily
 
 
 class DialogService(CommonService):
@@ -65,128 +62,51 @@ class DialogService(CommonService):
         return list(chats.dicts())
 
 
-def message_fit_in(msg, max_length=4000):
-    def count():
-        nonlocal msg
-        tks_cnts = []
-        for m in msg:
-            tks_cnts.append(
-                {"role": m["role"], "count": num_tokens_from_string(m["content"])})
-        total = 0
-        for m in tks_cnts:
-            total += m["count"]
-        return total
+def chat_solo(dialog, messages, stream=True):
+    if llm_id2llm_type(dialog.llm_id) == "image2text":
+        chat_mdl = LLMBundle(dialog.tenant_id, LLMType.IMAGE2TEXT, dialog.llm_id)
+    else:
+        chat_mdl = LLMBundle(dialog.tenant_id, LLMType.CHAT, dialog.llm_id)
 
-    c = count()
-    if c < max_length:
-        return c, msg
-
-    msg_ = [m for m in msg[:-1] if m["role"] == "system"]
-    if len(msg) > 1:
-        msg_.append(msg[-1])
-    msg = msg_
-    c = count()
-    if c < max_length:
-        return c, msg
-
-    ll = num_tokens_from_string(msg_[0]["content"])
-    ll2 = num_tokens_from_string(msg_[-1]["content"])
-    if ll / (ll + ll2) > 0.8:
-        m = msg_[0]["content"]
-        m = encoder.decode(encoder.encode(m)[:max_length - ll2])
-        msg[0]["content"] = m
-        return max_length, msg
-
-    m = msg_[1]["content"]
-    m = encoder.decode(encoder.encode(m)[:max_length - ll2])
-    msg[1]["content"] = m
-    return max_length, msg
-
-
-def llm_id2llm_type(llm_id):
-    llm_id, _ = TenantLLMService.split_model_name_and_factory(llm_id)
-    fnm = os.path.join(get_project_base_directory(), "conf")
-    llm_factories = json.load(open(os.path.join(fnm, "llm_factories.json"), "r"))
-    for llm_factory in llm_factories["factory_llm_infos"]:
-        for llm in llm_factory["llm"]:
-            if llm_id == llm["llm_name"]:
-                return llm["model_type"].strip(",")[-1]
-
-
-def kb_prompt(kbinfos, max_tokens):
-    knowledges = [ck["content_with_weight"] for ck in kbinfos["chunks"]]
-    used_token_count = 0
-    chunks_num = 0
-    for i, c in enumerate(knowledges):
-        used_token_count += num_tokens_from_string(c)
-        chunks_num += 1
-        if max_tokens * 0.97 < used_token_count:
-            knowledges = knowledges[:i]
-            break
-
-    docs = DocumentService.get_by_ids([ck["doc_id"] for ck in kbinfos["chunks"][:chunks_num]])
-    docs = {d.id: d.meta_fields for d in docs}
-
-    doc2chunks = defaultdict(lambda: {"chunks": [], "meta": []})
-    for ck in kbinfos["chunks"][:chunks_num]:
-        doc2chunks[ck["docnm_kwd"]]["chunks"].append(ck["content_with_weight"])
-        doc2chunks[ck["docnm_kwd"]]["meta"] = docs.get(ck["doc_id"], {})
-
-    knowledges = []
-    for nm, cks_meta in doc2chunks.items():
-        txt = f"Document: {nm} \n"
-        for k,v in cks_meta["meta"].items():
-            txt += f"{k}: {v}\n"
-        txt += "Relevant fragments as following:\n"
-        for i, chunk in enumerate(cks_meta["chunks"], 1):
-            txt += f"{i}. {chunk}\n"
-        knowledges.append(txt)
-    return knowledges
-
-
-def label_question(question, kbs):
-    tags = None
-    tag_kb_ids = []
-    for kb in kbs:
-        if kb.parser_config.get("tag_kb_ids"):
-            tag_kb_ids.extend(kb.parser_config["tag_kb_ids"])
-    if tag_kb_ids:
-        all_tags = get_tags_from_cache(tag_kb_ids)
-        if not all_tags:
-            all_tags = settings.retrievaler.all_tags_in_portion(kb.tenant_id, tag_kb_ids)
-            set_tags_to_cache(all_tags, tag_kb_ids)
-        else:
-            all_tags = json.loads(all_tags)
-        tag_kbs = KnowledgebaseService.get_by_ids(tag_kb_ids)
-        tags = settings.retrievaler.tag_query(question,
-                                              list(set([kb.tenant_id for kb in tag_kbs])),
-                                              tag_kb_ids,
-                                              all_tags,
-                                              kb.parser_config.get("topn_tags", 3)
-                                              )
-    return tags
+    prompt_config = dialog.prompt_config
+    tts_mdl = None
+    if prompt_config.get("tts"):
+        tts_mdl = LLMBundle(dialog.tenant_id, LLMType.TTS)
+    msg = [{"role": m["role"], "content": re.sub(r"##\d+\$\$", "", m["content"])}
+           for m in messages if m["role"] != "system"]
+    if stream:
+        last_ans = ""
+        for ans in chat_mdl.chat_streamly(prompt_config.get("system", ""), msg, dialog.llm_setting):
+            answer = ans
+            delta_ans = ans[len(last_ans):]
+            if num_tokens_from_string(delta_ans) < 16:
+                continue
+            last_ans = answer
+            yield {"answer": answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans), "prompt": "", "created_at": time.time()}
+        if delta_ans:
+            yield {"answer": answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans), "prompt": "", "created_at": time.time()}
+    else:
+        answer = chat_mdl.chat(prompt_config.get("system", ""), msg, dialog.llm_setting)
+        user_content = msg[-1].get("content", "[content not available]")
+        logging.debug("User: {}|Assistant: {}".format(user_content, answer))
+        yield {"answer": answer, "reference": {}, "audio_binary": tts(tts_mdl, answer), "prompt": "", "created_at": time.time()}
 
 
 def chat(dialog, messages, stream=True, **kwargs):
     assert messages[-1]["role"] == "user", "The last content of this conversation is not from user."
+    if not dialog.kb_ids:
+        for ans in chat_solo(dialog, messages, stream):
+            yield ans
+        return
 
     chat_start_ts = timer()
 
-    # Get llm model name and model provider name
-    llm_id, model_provider = TenantLLMService.split_model_name_and_factory(dialog.llm_id)
-
-    # Get llm model instance by model and provide name
-    llm = LLMService.query(llm_name=llm_id) if not model_provider else LLMService.query(llm_name=llm_id, fid=model_provider)
-
-    if not llm:
-        # Model name is provided by tenant, but not system built-in
-        llm = TenantLLMService.query(tenant_id=dialog.tenant_id, llm_name=llm_id) if not model_provider else \
-            TenantLLMService.query(tenant_id=dialog.tenant_id, llm_name=llm_id, llm_factory=model_provider)
-        if not llm:
-            raise LookupError("LLM(%s) not found" % dialog.llm_id)
-        max_tokens = 8192
+    if llm_id2llm_type(dialog.llm_id) == "image2text":
+        llm_model_config = TenantLLMService.get_model_config(dialog.tenant_id, LLMType.IMAGE2TEXT, dialog.llm_id)
     else:
-        max_tokens = llm[0].max_tokens
+        llm_model_config = TenantLLMService.get_model_config(dialog.tenant_id, LLMType.CHAT, dialog.llm_id)
+
+    max_tokens = llm_model_config.get("max_tokens", 8192)
 
     check_llm_ts = timer()
 
@@ -204,9 +124,6 @@ def chat(dialog, messages, stream=True, **kwargs):
     attachments = kwargs["doc_ids"].split(",") if "doc_ids" in kwargs else None
     if "doc_ids" in messages[-1]:
         attachments = messages[-1]["doc_ids"]
-        for m in messages[:-1]:
-            if "doc_ids" in m:
-                attachments.extend(m["doc_ids"])
 
     create_retriever_ts = timer()
 
@@ -258,9 +175,11 @@ def chat(dialog, messages, stream=True, **kwargs):
 
     bind_reranker_ts = timer()
     generate_keyword_ts = bind_reranker_ts
+    thought = ""
+    kbinfos = {"total": 0, "chunks": [], "doc_aggs": []}
 
     if "knowledge" not in [p["key"] for p in prompt_config["parameters"]]:
-        kbinfos = {"total": 0, "chunks": [], "doc_aggs": []}
+        knowledges = []
     else:
         if prompt_config.get("keyword", False):
             questions[-1] += keyword_extraction(chat_mdl, questions[-1])
@@ -268,43 +187,63 @@ def chat(dialog, messages, stream=True, **kwargs):
 
         tenant_ids = list(set([kb.tenant_id for kb in kbs]))
 
-        kbinfos = retriever.retrieval(" ".join(questions), embd_mdl, tenant_ids, dialog.kb_ids, 1, dialog.top_n,
-                                      dialog.similarity_threshold,
-                                      dialog.vector_similarity_weight,
-                                      doc_ids=attachments,
-                                      top=dialog.top_k, aggs=False, rerank_mdl=rerank_mdl,
-                                      rank_feature=label_question(" ".join(questions), kbs)
-                                      )
-        if prompt_config.get("use_kg"):
-            ck = settings.kg_retrievaler.retrieval(" ".join(questions),
-                                              tenant_ids,
-                                              dialog.kb_ids,
-                                              embd_mdl,
-                                              LLMBundle(dialog.tenant_id, LLMType.CHAT))
-            if ck["content_with_weight"]:
-                kbinfos["chunks"].insert(0, ck)
+        knowledges = []
+        if prompt_config.get("reasoning", False):
+            reasoner = DeepResearcher(chat_mdl,
+                                      prompt_config,
+                                      partial(retriever.retrieval, embd_mdl=embd_mdl, tenant_ids=tenant_ids, kb_ids=dialog.kb_ids, page=1, page_size=dialog.top_n, similarity_threshold=0.2, vector_similarity_weight=0.3))
 
-    retrieval_ts = timer()
+            for think in reasoner.thinking(kbinfos, " ".join(questions)):
+                if isinstance(think, str):
+                    thought = think
+                    knowledges = [t for t in think.split("\n") if t]
+                elif stream:
+                    yield think
+        else:
+            kbinfos = retriever.retrieval(" ".join(questions), embd_mdl, tenant_ids, dialog.kb_ids, 1, dialog.top_n,
+                                          dialog.similarity_threshold,
+                                          dialog.vector_similarity_weight,
+                                          doc_ids=attachments,
+                                          top=dialog.top_k, aggs=False, rerank_mdl=rerank_mdl,
+                                          rank_feature=label_question(" ".join(questions), kbs)
+                                          )
+            if prompt_config.get("tavily_api_key"):
+                tav = Tavily(prompt_config["tavily_api_key"])
+                tav_res = tav.retrieve_chunks(" ".join(questions))
+                kbinfos["chunks"].extend(tav_res["chunks"])
+                kbinfos["doc_aggs"].extend(tav_res["doc_aggs"])
+            if prompt_config.get("use_kg"):
+                ck = settings.kg_retrievaler.retrieval(" ".join(questions),
+                                                       tenant_ids,
+                                                       dialog.kb_ids,
+                                                       embd_mdl,
+                                                       LLMBundle(dialog.tenant_id, LLMType.CHAT))
+                if ck["content_with_weight"]:
+                    kbinfos["chunks"].insert(0, ck)
+
+            knowledges = kb_prompt(kbinfos, max_tokens)
 
-    knowledges = kb_prompt(kbinfos, max_tokens)
     logging.debug(
         "{}->{}".format(" ".join(questions), "\n->".join(knowledges)))
 
+    retrieval_ts = timer()
     if not knowledges and prompt_config.get("empty_response"):
         empty_res = prompt_config["empty_response"]
-        yield {"answer": empty_res, "reference": kbinfos, "audio_binary": tts(tts_mdl, empty_res)}
+        yield {"answer": empty_res, "reference": kbinfos, "prompt": "\n\n### Query:\n%s" % " ".join(questions), "audio_binary": tts(tts_mdl, empty_res)}
         return {"answer": prompt_config["empty_response"], "reference": kbinfos}
 
     kwargs["knowledge"] = "\n------\n" + "\n\n------\n\n".join(knowledges)
     gen_conf = dialog.llm_setting
 
     msg = [{"role": "system", "content": prompt_config["system"].format(**kwargs)}]
+    prompt4citation = ""
+    if knowledges and (prompt_config.get("quote", True) and kwargs.get("quote", True)):
+        prompt4citation = citation_prompt()
     msg.extend([{"role": m["role"], "content": re.sub(r"##\d+\$\$", "", m["content"])}
                 for m in messages if m["role"] != "system"])
-    used_token_count, msg = message_fit_in(msg, int(max_tokens * 0.97))
+    used_token_count, msg = message_fit_in(msg, int(max_tokens * 0.95))
     assert len(msg) >= 2, f"message_fit_in has bug: {msg}"
     prompt = msg[0]["content"]
-    prompt += "\n\n### Query:\n%s" % " ".join(questions)
 
     if "max_tokens" in gen_conf:
         gen_conf["max_tokens"] = min(
@@ -312,20 +251,32 @@ def chat(dialog, messages, stream=True, **kwargs):
             max_tokens - used_token_count)
 
     def decorate_answer(answer):
-        nonlocal prompt_config, knowledges, kwargs, kbinfos, prompt, retrieval_ts
-
-        finish_chat_ts = timer()
+        nonlocal prompt_config, knowledges, kwargs, kbinfos, prompt, retrieval_ts, questions
 
         refs = []
+        ans = answer.split("</think>")
+        think = ""
+        if len(ans) == 2:
+            think = ans[0] + "</think>"
+            answer = ans[1]
         if knowledges and (prompt_config.get("quote", True) and kwargs.get("quote", True)):
-            answer, idx = retriever.insert_citations(answer,
-                                                     [ck["content_ltks"]
-                                                      for ck in kbinfos["chunks"]],
-                                                     [ck["vector"]
-                                                      for ck in kbinfos["chunks"]],
-                                                     embd_mdl,
-                                                     tkweight=1 - dialog.vector_similarity_weight,
-                                                     vtweight=dialog.vector_similarity_weight)
+            answer = re.sub(r"##[ij]\$\$", "", answer, flags=re.DOTALL)
+            if not re.search(r"##[0-9]+\$\$", answer):
+                answer, idx = retriever.insert_citations(answer,
+                                                         [ck["content_ltks"]
+                                                          for ck in kbinfos["chunks"]],
+                                                         [ck["vector"]
+                                                          for ck in kbinfos["chunks"]],
+                                                         embd_mdl,
+                                                         tkweight=1 - dialog.vector_similarity_weight,
+                                                         vtweight=dialog.vector_similarity_weight)
+            else:
+                idx = set([])
+                for r in re.finditer(r"##([0-9]+)\$\$", answer):
+                    i = int(r.group(1))
+                    if i < len(kbinfos["chunks"]):
+                        idx.add(i)
+
             idx = set([kbinfos["chunks"][int(i)]["doc_id"] for i in idx])
             recall_docs = [
                 d for d in kbinfos["doc_aggs"] if d["doc_id"] in idx]
@@ -353,27 +304,30 @@ def chat(dialog, messages, stream=True, **kwargs):
         retrieval_time_cost = (retrieval_ts - generate_keyword_ts) * 1000
         generate_result_time_cost = (finish_chat_ts - retrieval_ts) * 1000
 
+        prompt += "\n\n### Query:\n%s" % " ".join(questions)
         prompt = f"{prompt}\n\n - Total: {total_time_cost:.1f}ms\n  - Check LLM: {check_llm_time_cost:.1f}ms\n  - Create retriever: {create_retriever_time_cost:.1f}ms\n  - Bind embedding: {bind_embedding_time_cost:.1f}ms\n  - Bind LLM: {bind_llm_time_cost:.1f}ms\n  - Tune question: {refine_question_time_cost:.1f}ms\n  - Bind reranker: {bind_reranker_time_cost:.1f}ms\n  - Generate keyword: {generate_keyword_time_cost:.1f}ms\n  - Retrieval: {retrieval_time_cost:.1f}ms\n  - Generate answer: {generate_result_time_cost:.1f}ms"
-        return {"answer": answer, "reference": refs, "prompt": re.sub(r"\n", "  \n", prompt)}
+        return {"answer": think+answer, "reference": refs, "prompt": re.sub(r"\n", "  \n", prompt), "created_at": time.time()}
 
     if stream:
         last_ans = ""
         answer = ""
-        for ans in chat_mdl.chat_streamly(prompt, msg[1:], gen_conf):
+        for ans in chat_mdl.chat_streamly(prompt+prompt4citation, msg[1:], gen_conf):
+            if thought:
+                ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
             answer = ans
             delta_ans = ans[len(last_ans):]
             if num_tokens_from_string(delta_ans) < 16:
                 continue
             last_ans = answer
-            yield {"answer": answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans)}
+            yield {"answer": thought+answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans)}
         delta_ans = answer[len(last_ans):]
         if delta_ans:
-            yield {"answer": answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans)}
-        yield decorate_answer(answer)
+            yield {"answer": thought+answer, "reference": {}, "audio_binary": tts(tts_mdl, delta_ans)}
+        yield decorate_answer(thought+answer)
     else:
-        answer = chat_mdl.chat(prompt, msg[1:], gen_conf)
-        logging.debug("User: {}|Assistant: {}".format(
-            msg[-1]["content"], answer))
+        answer = chat_mdl.chat(prompt+prompt4citation, msg[1:], gen_conf)
+        user_content = msg[-1].get("content", "[content not available]")
+        logging.debug("User: {}|Assistant: {}".format(user_content, answer))
         res = decorate_answer(answer)
         res["audio_binary"] = tts(tts_mdl, answer)
         yield res
@@ -400,6 +354,7 @@ Please write the SQL, only SQL, without any other explanations or text.
         nonlocal sys_prompt, user_prompt, question, tried_times
         sql = chat_mdl.chat(sys_prompt, [{"role": "user", "content": user_prompt}], {
             "temperature": 0.06})
+        sql = re.sub(r"<think>.*</think>", "", sql, flags=re.DOTALL)
         logging.debug(f"{question} ==> {user_prompt} get SQL: {sql}")
         sql = re.sub(r"[\r\n]+", " ", sql.lower())
         sql = re.sub(r".*select ", "select ", sql.lower())
@@ -506,172 +461,6 @@ Please write the SQL, only SQL, without any other explanations or text.
     }
 
 
-def relevant(tenant_id, llm_id, question, contents: list):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    prompt = """
-        You are a grader assessing relevance of a retrieved document to a user question. 
-        It does not need to be a stringent test. The goal is to filter out erroneous retrievals.
-        If the document contains keyword(s) or semantic meaning related to the user question, grade it as relevant. 
-        Give a binary score 'yes' or 'no' score to indicate whether the document is relevant to the question.
-        No other words needed except 'yes' or 'no'.
-    """
-    if not contents:
-        return False
-    contents = "Documents: \n" + "   - ".join(contents)
-    contents = f"Question: {question}\n" + contents
-    if num_tokens_from_string(contents) >= chat_mdl.max_length - 4:
-        contents = encoder.decode(encoder.encode(contents)[:chat_mdl.max_length - 4])
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": contents}], {"temperature": 0.01})
-    if ans.lower().find("yes") >= 0:
-        return True
-    return False
-
-
-def rewrite(tenant_id, llm_id, question):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    prompt = """
-        You are an expert at query expansion to generate a paraphrasing of a question.
-        I can't retrieval relevant information from the knowledge base by using user's question directly.     
-        You need to expand or paraphrase user's question by multiple ways such as using synonyms words/phrase, 
-        writing the abbreviation in its entirety, adding some extra descriptions or explanations, 
-        changing the way of expression, translating the original question into another language (English/Chinese), etc. 
-        And return 5 versions of question and one is from translation.
-        Just list the question. No other words are needed.
-    """
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": question}], {"temperature": 0.8})
-    return ans
-
-
-def keyword_extraction(chat_mdl, content, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-Task: extract the most important keywords/phrases of a given piece of text content.
-Requirements: 
-  - Summarize the text content, and give top {topn} important keywords/phrases.
-  - The keywords MUST be in language of the given piece of text content.
-  - The keywords are delimited by ENGLISH COMMA.
-  - Keywords ONLY in output.
-
-### Text Content 
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    if kwd.find("**ERROR**") >= 0:
-        return ""
-    return kwd
-
-
-def question_proposal(chat_mdl, content, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-Task:  propose {topn} questions about a given piece of text content.
-Requirements: 
-  - Understand and summarize the text content, and propose top {topn} important questions.
-  - The questions SHOULD NOT have overlapping meanings.
-  - The questions SHOULD cover the main content of the text as much as possible.
-  - The questions MUST be in language of the given piece of text content.
-  - One question per line.
-  - Question ONLY in output.
-
-### Text Content 
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    if kwd.find("**ERROR**") >= 0:
-        return ""
-    return kwd
-
-
-def full_question(tenant_id, llm_id, messages):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    conv = []
-    for m in messages:
-        if m["role"] not in ["user", "assistant"]:
-            continue
-        conv.append("{}: {}".format(m["role"].upper(), m["content"]))
-    conv = "\n".join(conv)
-    today = datetime.date.today().isoformat()
-    yesterday = (datetime.date.today() - timedelta(days=1)).isoformat()
-    tomorrow = (datetime.date.today() + timedelta(days=1)).isoformat()
-    prompt = f"""
-Role: A helpful assistant
-
-Task and steps: 
-    1. Generate a full user question that would follow the conversation.
-    2. If the user's question involves relative date, you need to convert it into absolute date based on the current date, which is {today}. For example: 'yesterday' would be converted to {yesterday}.
-    
-Requirements & Restrictions:
-  - Text generated MUST be in the same language of the original user's question.
-  - If the user's latest question is completely, don't do anything, just return the original question.
-  - DON'T generate anything except a refined question.
-
-######################
--Examples-
-######################
-
-# Example 1
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-###############
-Output: What's the name of Donald Trump's mother?
-
-------------
-# Example 2
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-ASSISTANT:  Mary Trump.
-User: What's her full name?
-###############
-Output: What's the full name of Donald Trump's mother Mary Trump?
-
-------------
-# Example 3
-## Conversation
-USER: What's the weather today in London?
-ASSISTANT:  Cloudy.
-USER: What's about tomorrow in Rochester?
-###############
-Output: What's the weather in Rochester on {tomorrow}?
-######################
-
-# Real Data
-## Conversation
-{conv}
-###############
-    """
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": "Output: "}], {"temperature": 0.2})
-    return ans if ans.find("**ERROR**") < 0 else messages[-1]["content"]
-
-
 def tts(tts_mdl, text):
     if not tts_mdl or not text:
         return
@@ -738,6 +527,7 @@ def ask(question, kb_ids, tenant_id):
 
         if answer.lower().find("invalid key") >= 0 or answer.lower().find("invalid api") >= 0:
             answer += " Please set LLM API-Key in 'User Setting -> Model Providers -> API-Key'"
+        refs["chunks"] = chunks_format(refs)
         return {"answer": answer, "reference": refs}
 
     answer = ""
@@ -745,64 +535,3 @@ def ask(question, kb_ids, tenant_id):
         answer = ans
         yield {"answer": answer, "reference": {}}
     yield decorate_answer(answer)
-
-
-def content_tagging(chat_mdl, content, all_tags, examples, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-
-Task: Tag (put on some labels) to a given piece of text content based on the examples and the entire tag set.
-
-Steps:: 
-  - Comprehend the tag/label set.
-  - Comprehend examples which all consist of both text content and assigned tags with relevance score in format of JSON.
-  - Summarize the text content, and tag it with top {topn} most relevant tags from the set of tag/label and the corresponding relevance score.
-
-Requirements
-  - The tags MUST be from the tag set.
-  - The output MUST be in JSON format only, the key is tag and the value is its relevance score.
-  - The relevance score must be range from 1 to 10.
-  - Keywords ONLY in output.
-
-# TAG SET
-{", ".join(all_tags)}
-
-"""
-    for i, ex in enumerate(examples):
-        prompt += """
-# Examples {}
-### Text Content
-{}
-
-Output:
-{}
-
-        """.format(i, ex["content"], json.dumps(ex[TAG_FLD], indent=2, ensure_ascii=False))
-
-    prompt += f"""
-# Real Data
-### Text Content
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.5})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    if kwd.find("**ERROR**") >= 0:
-        raise Exception(kwd)
-
-    try:
-        return json_repair.loads(kwd)
-    except json_repair.JSONDecodeError:
-        try:
-            result = kwd.replace(prompt[:-1], '').replace('user', '').replace('model', '').strip()
-            result = '{' + result.split('{')[1].split('}')[0] + '}'
-            return json_repair.loads(result)
-        except Exception as e:
-            logging.exception(f"JSON parsing error: {result} -> {e}")
-            raise e

api/db/services/document_service.py

@@ -22,13 +22,13 @@ from concurrent.futures import ThreadPoolExecutor
 from copy import deepcopy
 from datetime import datetime
 from io import BytesIO
+import trio
 
 from peewee import fn
 
 from api.db.db_utils import bulk_insert_into_db
 from api import settings
 from api.utils import current_timestamp, get_format_time, get_uuid
-from graphrag.general.mind_map_extractor import MindMapExtractor
 from rag.settings import SVR_QUEUE_NAME
 from rag.utils.storage_factory import STORAGE_IMPL
 from rag.nlp import search, rag_tokenizer
@@ -372,16 +372,14 @@ class DocumentService(CommonService):
                     "progress_msg": "Task is queued...",
                     "process_begin_at": get_format_time()
                     })
+    @classmethod
+    @DB.connection_context()
+    def update_meta_fields(cls, doc_id, meta_fields):
+        return cls.update_by_id(doc_id, {"meta_fields": meta_fields})
 
     @classmethod
     @DB.connection_context()
     def update_progress(cls):
-        MSG = {
-            "raptor": "Start RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval).",
-            "graphrag": "Start Graph Extraction",
-            "graph_resolution": "Start Graph Resolution",
-            "graph_community": "Start Graph Community Reports Generation"
-        }
         docs = cls.get_unfinished_docs()
         for d in docs:
             try:
@@ -392,37 +390,31 @@ class DocumentService(CommonService):
                 prg = 0
                 finished = True
                 bad = 0
+                has_raptor = False
+                has_graphrag = False
                 e, doc = DocumentService.get_by_id(d["id"])
                 status = doc.run  # TaskStatus.RUNNING.value
                 for t in tsks:
                     if 0 <= t.progress < 1:
                         finished = False
-                    prg += t.progress if t.progress >= 0 else 0
-                    if t.progress_msg not in msg:
-                        msg.append(t.progress_msg)
                     if t.progress == -1:
                         bad += 1
+                    prg += t.progress if t.progress >= 0 else 0
+                    msg.append(t.progress_msg)
+                    if t.task_type == "raptor":
+                        has_raptor = True
+                    elif t.task_type == "graphrag":
+                        has_graphrag = True
                 prg /= len(tsks)
                 if finished and bad:
                     prg = -1
                     status = TaskStatus.FAIL.value
                 elif finished:
-                    m = "\n".join(sorted(msg))
-                    if d["parser_config"].get("raptor", {}).get("use_raptor") and m.find(MSG["raptor"]) < 0:
-                        queue_raptor_o_graphrag_tasks(d, "raptor", MSG["raptor"])
+                    if d["parser_config"].get("raptor", {}).get("use_raptor") and not has_raptor:
+                        queue_raptor_o_graphrag_tasks(d, "raptor")
                         prg = 0.98 * len(tsks) / (len(tsks) + 1)
-                    elif d["parser_config"].get("graphrag", {}).get("use_graphrag") and m.find(MSG["graphrag"]) < 0:
-                        queue_raptor_o_graphrag_tasks(d, "graphrag", MSG["graphrag"])
-                        prg = 0.98 * len(tsks) / (len(tsks) + 1)
-                    elif d["parser_config"].get("graphrag", {}).get("use_graphrag") \
-                        and d["parser_config"].get("graphrag", {}).get("resolution") \
-                        and m.find(MSG["graph_resolution"]) < 0:
-                        queue_raptor_o_graphrag_tasks(d, "graph_resolution", MSG["graph_resolution"])
-                        prg = 0.98 * len(tsks) / (len(tsks) + 1)
-                    elif d["parser_config"].get("graphrag", {}).get("use_graphrag") \
-                        and d["parser_config"].get("graphrag", {}).get("community") \
-                        and m.find(MSG["graph_community"]) < 0:
-                        queue_raptor_o_graphrag_tasks(d, "graph_community", MSG["graph_community"])
+                    elif d["parser_config"].get("graphrag", {}).get("use_graphrag") and not has_graphrag:
+                        queue_raptor_o_graphrag_tasks(d, "graphrag")
                         prg = 0.98 * len(tsks) / (len(tsks) + 1)
                     else:
                         status = TaskStatus.DONE.value
@@ -459,7 +451,7 @@ class DocumentService(CommonService):
         return False
 
 
-def queue_raptor_o_graphrag_tasks(doc, ty, msg):
+def queue_raptor_o_graphrag_tasks(doc, ty):
     chunking_config = DocumentService.get_chunking_config(doc["id"])
     hasher = xxhash.xxh64()
     for field in sorted(chunking_config.keys()):
@@ -472,7 +464,8 @@ def queue_raptor_o_graphrag_tasks(doc, ty, msg):
             "doc_id": doc["id"],
             "from_page": 100000000,
             "to_page": 100000000,
-            "progress_msg":  datetime.now().strftime("%H:%M:%S") + " " + msg
+            "task_type": ty,
+            "progress_msg":  datetime.now().strftime("%H:%M:%S") + " created task " + ty
         }
 
     task = new_task()
@@ -481,7 +474,6 @@ def queue_raptor_o_graphrag_tasks(doc, ty, msg):
     hasher.update(ty.encode("utf-8"))
     task["digest"] = hasher.hexdigest()
     bulk_insert_into_db(Task, [task], True)
-    task["task_type"] = ty
     assert REDIS_CONN.queue_product(SVR_QUEUE_NAME, message=task), "Can't access Redis. Please check the Redis' status."
 
 
@@ -500,6 +492,9 @@ def doc_upload_and_parse(conversation_id, file_objs, user_id):
     assert e, "Conversation not found!"
 
     e, dia = DialogService.get_by_id(conv.dialog_id)
+    if not dia.kb_ids:
+        raise LookupError("No knowledge base associated with this conversation. "
+                          "Please add a knowledge base before uploading documents")
     kb_id = dia.kb_ids[0]
     e, kb = KnowledgebaseService.get_by_id(kb_id)
     if not e:
@@ -588,10 +583,11 @@ def doc_upload_and_parse(conversation_id, file_objs, user_id):
         cks = [c for c in docs if c["doc_id"] == doc_id]
 
         if parser_ids[doc_id] != ParserType.PICTURE.value:
+            from graphrag.general.mind_map_extractor import MindMapExtractor
             mindmap = MindMapExtractor(llm_bdl)
             try:
-                mind_map = json.dumps(mindmap([c["content_with_weight"] for c in docs if c["doc_id"] == doc_id]).output,
-                                      ensure_ascii=False, indent=2)
+                mind_map = trio.run(mindmap, [c["content_with_weight"] for c in docs if c["doc_id"] == doc_id])
+                mind_map = json.dumps(mind_map.output, ensure_ascii=False, indent=2)
                 if len(mind_map) < 32:
                     raise Exception("Few content: " + mind_map)
                 cks.append({

api/db/services/knowledgebase_service.py

@@ -22,6 +22,42 @@ from peewee import fn
 class KnowledgebaseService(CommonService):
     model = Knowledgebase
 
+    @classmethod
+    @DB.connection_context()
+    def is_parsed_done(cls, kb_id):
+        """
+        Check if all documents in the knowledge base have completed parsing
+        
+        Args:
+            kb_id: Knowledge base ID
+            
+        Returns:
+            If all documents are parsed successfully, returns (True, None)
+            If any document is not fully parsed, returns (False, error_message)
+        """
+        from api.db import TaskStatus
+        from api.db.services.document_service import DocumentService
+        
+        # Get knowledge base information
+        kbs = cls.query(id=kb_id)
+        if not kbs:
+            return False, "Knowledge base not found"
+        kb = kbs[0]
+        
+        # Get all documents in the knowledge base
+        docs, _ = DocumentService.get_by_kb_id(kb_id, 1, 1000, "create_time", True, "")
+        
+        # Check parsing status of each document
+        for doc in docs:
+            # If document is being parsed, don't allow chat creation
+            if doc['run'] == TaskStatus.RUNNING.value or doc['run'] == TaskStatus.CANCEL.value or doc['run'] == TaskStatus.FAIL.value:
+                return False, f"Document '{doc['name']}' in dataset '{kb.name}' is still being parsed. Please wait until all documents are parsed before starting a chat."
+            # If document is not yet parsed and has no chunks, don't allow chat creation
+            if doc['run'] == TaskStatus.UNSTART.value and doc['chunk_num'] == 0:
+                return False, f"Document '{doc['name']}' in dataset '{kb.name}' has not been parsed yet. Please parse all documents before starting a chat."
+        
+        return True, None
+
     @classmethod
     @DB.connection_context()
     def list_documents_by_ids(cls,kb_ids):

api/db/services/llm_service.py

@@ -86,8 +86,7 @@ class TenantLLMService(CommonService):
 
     @classmethod
     @DB.connection_context()
-    def model_instance(cls, tenant_id, llm_type,
-                       llm_name=None, lang="Chinese"):
+    def get_model_config(cls, tenant_id, llm_type, llm_name=None):
         e, tenant = TenantService.get_by_id(tenant_id)
         if not e:
             raise LookupError("Tenant not found")
@@ -124,7 +123,13 @@ class TenantLLMService(CommonService):
                     if not mdlnm:
                         raise LookupError(f"Type of {llm_type} model is not set.")
                     raise LookupError("Model({}) not authorized".format(mdlnm))
+        return model_config
 
+    @classmethod
+    @DB.connection_context()
+    def model_instance(cls, tenant_id, llm_type,
+                       llm_name=None, lang="Chinese"):
+        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
         if llm_type == LLMType.EMBEDDING.value:
             if model_config["llm_factory"] not in EmbeddingModel:
                 return
@@ -173,40 +178,39 @@ class TenantLLMService(CommonService):
     def increase_usage(cls, tenant_id, llm_type, used_tokens, llm_name=None):
         e, tenant = TenantService.get_by_id(tenant_id)
         if not e:
-            raise LookupError("Tenant not found")
+            logging.error(f"Tenant not found: {tenant_id}")
+            return 0
 
-        if llm_type == LLMType.EMBEDDING.value:
-            mdlnm = tenant.embd_id
-        elif llm_type == LLMType.SPEECH2TEXT.value:
-            mdlnm = tenant.asr_id
-        elif llm_type == LLMType.IMAGE2TEXT.value:
-            mdlnm = tenant.img2txt_id
-        elif llm_type == LLMType.CHAT.value:
-            mdlnm = tenant.llm_id if not llm_name else llm_name
-        elif llm_type == LLMType.RERANK:
-            mdlnm = tenant.rerank_id if not llm_name else llm_name
-        elif llm_type == LLMType.TTS:
-            mdlnm = tenant.tts_id if not llm_name else llm_name
-        else:
-            assert False, "LLM type error"
+        llm_map = {
+            LLMType.EMBEDDING.value: tenant.embd_id,
+            LLMType.SPEECH2TEXT.value: tenant.asr_id,
+            LLMType.IMAGE2TEXT.value: tenant.img2txt_id,
+            LLMType.CHAT.value: tenant.llm_id if not llm_name else llm_name,
+            LLMType.RERANK.value: tenant.rerank_id if not llm_name else llm_name,
+            LLMType.TTS.value: tenant.tts_id if not llm_name else llm_name
+        }
+
+        mdlnm = llm_map.get(llm_type)
+        if mdlnm is None:
+            logging.error(f"LLM type error: {llm_type}")
+            return 0
 
         llm_name, llm_factory = TenantLLMService.split_model_name_and_factory(mdlnm)
 
-        num = 0
         try:
-            if llm_factory:
-                tenant_llms = cls.query(tenant_id=tenant_id, llm_name=llm_name, llm_factory=llm_factory)
-            else:
-                tenant_llms = cls.query(tenant_id=tenant_id, llm_name=llm_name)
-            if not tenant_llms:
-                return num
-            else:
-                tenant_llm = tenant_llms[0]
-                num = cls.model.update(used_tokens=tenant_llm.used_tokens + used_tokens) \
-                    .where(cls.model.tenant_id == tenant_id, cls.model.llm_factory == tenant_llm.llm_factory, cls.model.llm_name == llm_name) \
-                    .execute()
+            num = cls.model.update(
+                used_tokens=cls.model.used_tokens + used_tokens
+            ).where(
+                cls.model.tenant_id == tenant_id,
+                cls.model.llm_name == llm_name,
+                cls.model.llm_factory == llm_factory if llm_factory else True
+            ).execute()
         except Exception:
-            logging.exception("TenantLLMService.increase_usage got exception")
+            logging.exception(
+                "TenantLLMService.increase_usage got exception,Failed to update used_tokens for tenant_id=%s, llm_name=%s",
+                tenant_id, llm_name)
+            return 0
+
         return num
 
     @classmethod
@@ -220,7 +224,7 @@ class TenantLLMService(CommonService):
         return list(objs)
 
 
-class LLMBundle(object):
+class LLMBundle:
     def __init__(self, tenant_id, llm_type, llm_name=None, lang="Chinese"):
         self.tenant_id = tenant_id
         self.llm_type = llm_type
@@ -229,10 +233,8 @@ class LLMBundle(object):
             tenant_id, llm_type, llm_name, lang=lang)
         assert self.mdl, "Can't find model for {}/{}/{}".format(
             tenant_id, llm_type, llm_name)
-        self.max_length = 8192
-        for lm in LLMService.query(llm_name=llm_name):
-            self.max_length = lm.max_tokens
-            break
+        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
+        self.max_length = model_config.get("max_tokens", 8192)
 
     def encode(self, texts: list):
         embeddings, used_tokens = self.mdl.encode(texts)

api/ragflow_server.py

@@ -28,6 +28,7 @@ import sys
 import time
 import traceback
 from concurrent.futures import ThreadPoolExecutor
+import threading
 
 from werkzeug.serving import run_simple
 from api import settings
@@ -41,16 +42,28 @@ from api.db.init_data import init_web_data
 from api.versions import get_ragflow_version
 from api.utils import show_configs
 from rag.settings import print_rag_settings
+from rag.utils.redis_conn import RedisDistributedLock
 
+stop_event = threading.Event()
 
 def update_progress():
-    while True:
-        time.sleep(6)
+    redis_lock = RedisDistributedLock("update_progress", timeout=60)
+    while not stop_event.is_set():
         try:
+            if not redis_lock.acquire():
+                continue
             DocumentService.update_progress()
+            stop_event.wait(6)
         except Exception:
             logging.exception("update_progress exception")
+        finally:
+            redis_lock.release()
 
+def signal_handler(sig, frame):
+    logging.info("Received interrupt signal, shutting down...")
+    stop_event.set()
+    time.sleep(1)
+    sys.exit(0)
 
 if __name__ == '__main__':
     logging.info(r"""
@@ -96,6 +109,9 @@ if __name__ == '__main__':
     RuntimeConfig.init_env()
     RuntimeConfig.init_config(JOB_SERVER_HOST=settings.HOST_IP, HTTP_PORT=settings.HOST_PORT)
 
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
     thread = ThreadPoolExecutor(max_workers=1)
     thread.submit(update_progress)
 
@@ -112,4 +128,6 @@ if __name__ == '__main__':
         )
     except Exception:
         traceback.print_exc()
+        stop_event.set()
+        time.sleep(1)
         os.kill(os.getpid(), signal.SIGKILL)

api/settings.py

@@ -66,75 +66,28 @@ def init_settings():
     DATABASE_TYPE = os.getenv("DB_TYPE", 'mysql')
     DATABASE = decrypt_database_config(name=DATABASE_TYPE)
     LLM = get_base_config("user_default_llm", {})
+    LLM_DEFAULT_MODELS = LLM.get("default_models", {})
     LLM_FACTORY = LLM.get("factory", "Tongyi-Qianwen")
     LLM_BASE_URL = LLM.get("base_url")
 
     global CHAT_MDL, EMBEDDING_MDL, RERANK_MDL, ASR_MDL, IMAGE2TEXT_MDL
     if not LIGHTEN:
-        default_llm = {
-            "Tongyi-Qianwen": {
-                "chat_model": "qwen-plus",
-                "embedding_model": "text-embedding-v2",
-                "image2text_model": "qwen-vl-max",
-                "asr_model": "paraformer-realtime-8k-v1",
-            },
-            "OpenAI": {
-                "chat_model": "gpt-3.5-turbo",
-                "embedding_model": "text-embedding-ada-002",
-                "image2text_model": "gpt-4-vision-preview",
-                "asr_model": "whisper-1",
-            },
-            "Azure-OpenAI": {
-                "chat_model": "gpt-35-turbo",
-                "embedding_model": "text-embedding-ada-002",
-                "image2text_model": "gpt-4-vision-preview",
-                "asr_model": "whisper-1",
-            },
-            "ZHIPU-AI": {
-                "chat_model": "glm-3-turbo",
-                "embedding_model": "embedding-2",
-                "image2text_model": "glm-4v",
-                "asr_model": "",
-            },
-            "Ollama": {
-                "chat_model": "qwen-14B-chat",
-                "embedding_model": "flag-embedding",
-                "image2text_model": "",
-                "asr_model": "",
-            },
-            "Moonshot": {
-                "chat_model": "moonshot-v1-8k",
-                "embedding_model": "",
-                "image2text_model": "",
-                "asr_model": "",
-            },
-            "DeepSeek": {
-                "chat_model": "deepseek-chat",
-                "embedding_model": "",
-                "image2text_model": "",
-                "asr_model": "",
-            },
-            "VolcEngine": {
-                "chat_model": "",
-                "embedding_model": "",
-                "image2text_model": "",
-                "asr_model": "",
-            },
-            "BAAI": {
-                "chat_model": "",
-                "embedding_model": "BAAI/bge-large-zh-v1.5",
-                "image2text_model": "",
-                "asr_model": "",
-                "rerank_model": "BAAI/bge-reranker-v2-m3",
-            }
-        }
+        EMBEDDING_MDL = "BAAI/bge-large-zh-v1.5@BAAI"
 
-        if LLM_FACTORY:
-            CHAT_MDL = default_llm[LLM_FACTORY]["chat_model"] + f"@{LLM_FACTORY}"
-            ASR_MDL = default_llm[LLM_FACTORY]["asr_model"] + f"@{LLM_FACTORY}"
-            IMAGE2TEXT_MDL = default_llm[LLM_FACTORY]["image2text_model"] + f"@{LLM_FACTORY}"
-        EMBEDDING_MDL = default_llm["BAAI"]["embedding_model"] + "@BAAI"
-        RERANK_MDL = default_llm["BAAI"]["rerank_model"] + "@BAAI"
+    if LLM_DEFAULT_MODELS:
+        CHAT_MDL = LLM_DEFAULT_MODELS.get("chat_model", CHAT_MDL)
+        EMBEDDING_MDL = LLM_DEFAULT_MODELS.get("embedding_model", EMBEDDING_MDL)
+        RERANK_MDL = LLM_DEFAULT_MODELS.get("rerank_model", RERANK_MDL)
+        ASR_MDL = LLM_DEFAULT_MODELS.get("asr_model", ASR_MDL)
+        IMAGE2TEXT_MDL = LLM_DEFAULT_MODELS.get("image2text_model", IMAGE2TEXT_MDL)
+
+        # factory can be specified in the config name with "@". LLM_FACTORY will be used if not specified
+        CHAT_MDL = CHAT_MDL + (f"@{LLM_FACTORY}" if "@" not in CHAT_MDL and CHAT_MDL != "" else "")
+        EMBEDDING_MDL = EMBEDDING_MDL + (f"@{LLM_FACTORY}" if "@" not in EMBEDDING_MDL and EMBEDDING_MDL != "" else "")
+        RERANK_MDL = RERANK_MDL + (f"@{LLM_FACTORY}" if "@" not in RERANK_MDL and RERANK_MDL != "" else "")
+        ASR_MDL = ASR_MDL + (f"@{LLM_FACTORY}" if "@" not in ASR_MDL and ASR_MDL != "" else "")
+        IMAGE2TEXT_MDL = IMAGE2TEXT_MDL + (
+            f"@{LLM_FACTORY}" if "@" not in IMAGE2TEXT_MDL and IMAGE2TEXT_MDL != "" else "")
 
     global API_KEY, PARSERS, HOST_IP, HOST_PORT, SECRET_KEY
     API_KEY = LLM.get("api_key", "")

api/utils/__init__.py

@@ -70,6 +70,12 @@ def show_configs():
             if "password" in v:
                 v = copy.deepcopy(v)
                 v["password"] = "*" * 8
+            if "access_key" in v:
+                v = copy.deepcopy(v)
+                v["access_key"] = "*" * 8
+            if "secret_key" in v:
+                v = copy.deepcopy(v)
+                v["secret_key"] = "*" * 8
         msg += f"\n\t{k}: {v}"
     logging.info(msg)
 
@@ -351,6 +357,26 @@ def decrypt(line):
         line), "Fail to decrypt password!").decode('utf-8')
 
 
+def decrypt2(crypt_text):
+    from base64 import b64decode, b16decode
+    from Crypto.Cipher import PKCS1_v1_5 as Cipher_PKCS1_v1_5
+    from Crypto.PublicKey import RSA
+    decode_data = b64decode(crypt_text)
+    if len(decode_data) == 127:
+        hex_fixed = '00' + decode_data.hex()
+        decode_data = b16decode(hex_fixed.upper())
+
+    file_path = os.path.join(
+        file_utils.get_project_base_directory(),
+        "conf",
+        "private.pem")
+    pem = open(file_path).read()
+    rsa_key = RSA.importKey(pem, "Welcome")
+    cipher = Cipher_PKCS1_v1_5.new(rsa_key)
+    decrypt_text = cipher.decrypt(decode_data, None)
+    return (b64decode(decrypt_text)).decode()
+
+
 def download_img(url):
     if not url:
         return ""

api/utils/api_utils.py

@@ -335,11 +335,9 @@ def generate_confirmation_token(tenent_id):
     return "ragflow-" + serializer.dumps(get_uuid(), salt=tenent_id)[2:34]
 
 
-def valid(permission, valid_permission, language, valid_language, chunk_method, valid_chunk_method):
+def valid(permission, valid_permission, chunk_method, valid_chunk_method):
     if valid_parameter(permission, valid_permission):
         return valid_parameter(permission, valid_permission)
-    if valid_parameter(language, valid_language):
-        return valid_parameter(language, valid_language)
     if valid_parameter(chunk_method, valid_chunk_method):
         return valid_parameter(chunk_method, valid_chunk_method)
 
@@ -373,3 +371,32 @@ def get_parser_config(chunk_method, parser_config):
         "picture": None}
     parser_config = key_mapping[chunk_method]
     return parser_config
+
+
+def valid_parser_config(parser_config):
+    if not parser_config:
+        return
+    scopes = set([
+        "chunk_token_num",
+        "delimiter",
+        "raptor",
+        "graphrag",
+        "layout_recognize",
+        "task_page_size",
+        "pages",
+        "html4excel",
+        "auto_keywords",
+        "auto_questions",
+        "tag_kb_ids",
+        "topn_tags",
+        "filename_embd_weight"
+    ])
+    for k in parser_config.keys():
+        assert k in scopes, f"Abnormal 'parser_config'. Invalid key: {k}"
+
+    assert 1 <= parser_config.get("chunk_token_num", 1) < 100000000, "chunk_token_num should be in range from 1 to 100000000"
+    assert 1 <= parser_config.get("task_page_size", 1) < 100000000, "task_page_size should be in range from 1 to 100000000"
+    assert 0 <= parser_config.get("auto_keywords", 0) < 32, "auto_keywords should be in range from 0 to 32"
+    assert 0 <= parser_config.get("auto_questions", 0) < 10, "auto_questions should be in range from 0 to 10"
+    assert 0 <= parser_config.get("topn_tags", 0) < 10, "topn_tags should be in range from 0 to 10"
+    assert isinstance(parser_config.get("html4excel", False), bool), "html4excel should be True or False"

api/utils/file_utils.py

@@ -17,6 +17,8 @@ import base64
 import json
 import os
 import re
+import sys
+import threading
 from io import BytesIO
 
 import pdfplumber
@@ -30,6 +32,10 @@ from api.constants import IMG_BASE64_PREFIX
 PROJECT_BASE = os.getenv("RAG_PROJECT_BASE") or os.getenv("RAG_DEPLOY_BASE")
 RAG_BASE = os.getenv("RAG_BASE")
 
+LOCK_KEY_pdfplumber = "global_shared_lock_pdfplumber"
+if LOCK_KEY_pdfplumber not in sys.modules:
+    sys.modules[LOCK_KEY_pdfplumber] = threading.Lock()
+
 
 def get_project_base_directory(*args):
     global PROJECT_BASE
@@ -175,19 +181,21 @@ def thumbnail_img(filename, blob):
     """
     filename = filename.lower()
     if re.match(r".*\.pdf$", filename):
-        pdf = pdfplumber.open(BytesIO(blob))
-        buffered = BytesIO()
-        resolution = 32
-        img = None
-        for _ in range(10):
-            # https://github.com/jsvine/pdfplumber?tab=readme-ov-file#creating-a-pageimage-with-to_image
-            pdf.pages[0].to_image(resolution=resolution).annotated.save(buffered, format="png")
-            img = buffered.getvalue()
-            if len(img) >= 64000 and resolution >= 2:
-                resolution = resolution / 2
-                buffered = BytesIO()
-            else:
-                break
+        with sys.modules[LOCK_KEY_pdfplumber]:
+            pdf = pdfplumber.open(BytesIO(blob))
+            buffered = BytesIO()
+            resolution = 32
+            img = None
+            for _ in range(10):
+                # https://github.com/jsvine/pdfplumber?tab=readme-ov-file#creating-a-pageimage-with-to_image
+                pdf.pages[0].to_image(resolution=resolution).annotated.save(buffered, format="png")
+                img = buffered.getvalue()
+                if len(img) >= 64000 and resolution >= 2:
+                    resolution = resolution / 2
+                    buffered = BytesIO()
+                else:
+                    break
+        pdf.close()
         return img
 
     elif re.match(r".*\.(jpg|jpeg|png|tif|gif|icon|ico|webp)$", filename):

api/utils/log_utils.py

@@ -18,6 +18,8 @@ import os.path
 import logging
 from logging.handlers import RotatingFileHandler
 
+initialized_root_logger = False
+
 def get_project_base_directory():
     PROJECT_BASE = os.path.abspath(
         os.path.join(
@@ -29,10 +31,13 @@ def get_project_base_directory():
     return PROJECT_BASE
 
 def initRootLogger(logfile_basename: str, log_format: str = "%(asctime)-15s %(levelname)-8s %(process)d %(message)s"):
-    logger = logging.getLogger()
-    if logger.hasHandlers():
+    global initialized_root_logger
+    if initialized_root_logger:
         return
+    initialized_root_logger = True
 
+    logger = logging.getLogger()
+    logger.handlers.clear()
     log_path = os.path.abspath(os.path.join(get_project_base_directory(), "logs", f"{logfile_basename}.log"))
 
     os.makedirs(os.path.dirname(log_path), exist_ok=True)

conf/infinity_mapping.json

@@ -9,10 +9,10 @@
 	"title_tks": {"type": "varchar", "default": "", "analyzer": "whitespace"},
 	"title_sm_tks": {"type": "varchar", "default": "", "analyzer": "whitespace"},
 	"name_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
-	"important_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
-	"tag_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
+	"important_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace-#"},
+	"tag_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace-#"},
 	"important_tks": {"type": "varchar", "default": "", "analyzer": "whitespace"},
-	"question_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
+	"question_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace-#"},
 	"question_tks": {"type": "varchar", "default": "", "analyzer": "whitespace"},
 	"content_with_weight": {"type": "varchar", "default": ""},
 	"content_ltks": {"type": "varchar", "default": "", "analyzer": "whitespace"},
@@ -28,7 +28,7 @@
 	"rank_flt": {"type": "float", "default": 0},
 	"available_int": {"type": "integer", "default": 1},
 	"knowledge_graph_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
-	"entities_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace"},
+	"entities_kwd": {"type": "varchar", "default": "", "analyzer": "whitespace-#"},
 	"pagerank_fea": {"type": "integer", "default":  0},
 	"tag_feas": {"type": "varchar", "default":  ""},
 

conf/service_conf.yaml

@@ -5,25 +5,25 @@ mysql:
   name: 'rag_flow'
   user: 'root'
   password: 'infini_rag_flow'
-  host: 'mysql'
+  host: 'localhost'
   port: 5455
   max_connections: 100
   stale_timeout: 30
 minio:
   user: 'rag_flow'
   password: 'infini_rag_flow'
-  host: 'minio:9000'
+  host: 'localhost:9000'
 es:
-  hosts: 'http://es01:1200'
+  hosts: 'http://localhost:1200'
   username: 'elastic'
   password: 'infini_rag_flow'
 infinity:
-  uri: 'infinity:23817'
+  uri: 'localhost:23817'
   db_name: 'default_db'
 redis:
   db: 1
   password: 'infini_rag_flow'
-  host: 'redis:6379'
+  host: 'localhost:6379'
 
 # postgres:
 #   name: 'rag_flow'
@@ -37,6 +37,12 @@ redis:
 #   access_key: 'access_key'
 #   secret_key: 'secret_key'
 #   region: 'region'
+# oss:
+#   access_key: 'access_key'
+#   secret_key: 'secret_key'
+#   endpoint_url: 'http://oss-cn-hangzhou.aliyuncs.com'
+#   region: 'cn-hangzhou'
+#   bucket: 'bucket_name'
 # azure:
 #   auth_type: 'sas'
 #   container_url: 'container_url'

deepdoc/README_zh.md

@@ -113,4 +113,4 @@ PDF、DOCX、EXCEL和PPT四种文档格式都有相应的解析器。最复杂
   
 ### 简历
 
-简历是一种非常复杂的文件。一份由各种布局的非结构化文本组成的简历可以分解为由近百个字段组成的结构化数据。我们还没有打开解析器，因为我们在解析过程之后打开了处理方法。
+简历是一种非常复杂的文档。由各种格式的非结构化文本构成的简历可以被解析为包含近百个字段的结构化数据。我们还没有启用解析器，因为在解析过程之后才会启动处理方法。

deepdoc/parser/excel_parser.py

@@ -1,6 +1,3 @@
-#
-#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
-#
 #  Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 #  You may obtain a copy of the License at
@@ -14,20 +11,68 @@
 #  limitations under the License.
 #
 
-from openpyxl import load_workbook
+import logging
 import sys
 from io import BytesIO
 
+import pandas as pd
+from openpyxl import Workbook, load_workbook
+
 from rag.nlp import find_codec
 
 
 class RAGFlowExcelParser:
-    def html(self, fnm, chunk_rows=256):
-        if isinstance(fnm, str):
-            wb = load_workbook(fnm)
-        else:
-            wb = load_workbook(BytesIO(fnm))
 
+    @staticmethod
+    def _load_excel_to_workbook(file_like_object):
+        if isinstance(file_like_object, bytes):
+            file_like_object = BytesIO(file_like_object)
+
+        # Read first 4 bytes to determine file type
+        file_like_object.seek(0)
+        file_head = file_like_object.read(4)
+        file_like_object.seek(0)
+
+        if not (file_head.startswith(b'PK\x03\x04') or file_head.startswith(b'\xD0\xCF\x11\xE0')):
+            logging.info("****wxy: Not an Excel file, converting CSV to Excel Workbook")
+
+            try:
+                file_like_object.seek(0)
+                df = pd.read_csv(file_like_object)
+                return RAGFlowExcelParser._dataframe_to_workbook(df)
+
+            except Exception as e_csv:
+                raise Exception(f"****wxy: Failed to parse CSV and convert to Excel Workbook: {e_csv}")
+
+        try:
+            return load_workbook(file_like_object)
+        except Exception as e:
+            logging.info(f"****wxy: openpyxl load error: {e}, try pandas instead")
+            try:
+                file_like_object.seek(0)
+                df = pd.read_excel(file_like_object)
+                return RAGFlowExcelParser._dataframe_to_workbook(df)
+            except Exception as e_pandas:
+                raise Exception(f"****wxy: pandas.read_excel error: {e_pandas}, original openpyxl error: {e}")
+
+    @staticmethod
+    def _dataframe_to_workbook(df):
+        wb = Workbook()
+        ws = wb.active
+        ws.title = "Data"
+
+        for col_num, column_name in enumerate(df.columns, 1):
+            ws.cell(row=1, column=col_num, value=column_name)
+
+        for row_num, row in enumerate(df.values, 2):
+            for col_num, value in enumerate(row, 1):
+                ws.cell(row=row_num, column=col_num, value=value)
+
+        return wb
+
+    def html(self, fnm, chunk_rows=256):
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
         tb_chunks = []
         for sheetname in wb.sheetnames:
             ws = wb[sheetname]
@@ -45,7 +90,7 @@ class RAGFlowExcelParser:
                 tb += f"<table><caption>{sheetname}</caption>"
                 tb += tb_rows_0
                 for r in list(
-                    rows[1 + chunk_i * chunk_rows : 1 + (chunk_i + 1) * chunk_rows]
+                    rows[1 + chunk_i * chunk_rows: 1 + (chunk_i + 1) * chunk_rows]
                 ):
                     tb += "<tr>"
                     for i, c in enumerate(r):
@@ -60,10 +105,9 @@ class RAGFlowExcelParser:
         return tb_chunks
 
     def __call__(self, fnm):
-        if isinstance(fnm, str):
-            wb = load_workbook(fnm)
-        else:
-            wb = load_workbook(BytesIO(fnm))
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
+
         res = []
         for sheetname in wb.sheetnames:
             ws = wb[sheetname]
@@ -88,7 +132,7 @@ class RAGFlowExcelParser:
     @staticmethod
     def row_number(fnm, binary):
         if fnm.split(".")[-1].lower().find("xls") >= 0:
-            wb = load_workbook(BytesIO(binary))
+            wb = RAGFlowExcelParser._load_excel_to_workbook(BytesIO(binary))
             total = 0
             for sheetname in wb.sheetnames:
                 ws = wb[sheetname]

deepdoc/parser/markdown_parser.py

@@ -22,27 +22,56 @@ class RAGFlowMarkdownParser:
         self.chunk_token_num = int(chunk_token_num)
 
     def extract_tables_and_remainder(self, markdown_text):
-        # Standard Markdown table
-        table_pattern = re.compile(
-            r'''
-            (?:\n|^)                     
-            (?:\|.*?\|.*?\|.*?\n)        
-            (?:\|(?:\s*[:-]+[-| :]*\s*)\|.*?\n) 
-            (?:\|.*?\|.*?\|.*?\n)+
+        tables = []
+        remainder = markdown_text
+        if "|" in markdown_text: # for optimize performance
+            # Standard Markdown table
+            border_table_pattern = re.compile(
+                r'''
+                (?:\n|^)                     
+                (?:\|.*?\|.*?\|.*?\n)        
+                (?:\|(?:\s*[:-]+[-| :]*\s*)\|.*?\n) 
+                (?:\|.*?\|.*?\|.*?\n)+
             ''', re.VERBOSE)
-        tables = table_pattern.findall(markdown_text)
-        remainder = table_pattern.sub('', markdown_text)
+            border_tables = border_table_pattern.findall(markdown_text)
+            tables.extend(border_tables)
+            remainder = border_table_pattern.sub('', remainder)
 
-        # Borderless Markdown table
-        no_border_table_pattern = re.compile(
+            # Borderless Markdown table
+            no_border_table_pattern = re.compile(
+                r'''
+                (?:\n|^)                 
+                (?:\S.*?\|.*?\n)
+                (?:(?:\s*[:-]+[-| :]*\s*).*?\n)
+                (?:\S.*?\|.*?\n)+
+                ''', re.VERBOSE)
+            no_border_tables = no_border_table_pattern.findall(remainder)
+            tables.extend(no_border_tables)
+            remainder = no_border_table_pattern.sub('', remainder)
+
+        if "<table>" in remainder.lower(): # for optimize performance
+            #HTML table extraction - handle possible html/body wrapper tags
+            html_table_pattern = re.compile(
             r'''
-            (?:\n|^)                 
-            (?:\S.*?\|.*?\n)
-            (?:(?:\s*[:-]+[-| :]*\s*).*?\n)
-            (?:\S.*?\|.*?\n)+
-            ''', re.VERBOSE)
-        no_border_tables = no_border_table_pattern.findall(remainder)
-        tables.extend(no_border_tables)
-        remainder = no_border_table_pattern.sub('', remainder)
+            (?:\n|^)
+            \s*
+            (?:
+                # case1: <html><body><table>...</table></body></html>
+                (?:<html[^>]*>\s*<body[^>]*>\s*<table[^>]*>.*?</table>\s*</body>\s*</html>)
+                |
+                # case2: <body><table>...</table></body>
+                (?:<body[^>]*>\s*<table[^>]*>.*?</table>\s*</body>)
+                |
+                # case3: only<table>...</table>
+                (?:<table[^>]*>.*?</table>)
+            )
+            \s*
+            (?=\n|$)
+            ''',
+            re.VERBOSE | re.DOTALL | re.IGNORECASE
+            )
+            html_tables = html_table_pattern.findall(remainder)
+            tables.extend(html_tables)
+            remainder = html_table_pattern.sub('', remainder)
 
         return remainder, tables

deepdoc/parser/pdf_parser.py

@@ -17,6 +17,9 @@
 import logging
 import os
 import random
+from timeit import default_timer as timer
+import sys
+import threading
 
 import xgboost as xgb
 from io import BytesIO
@@ -33,8 +36,23 @@ from rag.nlp import rag_tokenizer
 from copy import deepcopy
 from huggingface_hub import snapshot_download
 
+LOCK_KEY_pdfplumber = "global_shared_lock_pdfplumber"
+if LOCK_KEY_pdfplumber not in sys.modules:
+    sys.modules[LOCK_KEY_pdfplumber] = threading.Lock()
+
 class RAGFlowPdfParser:
     def __init__(self):
+        """
+        If you have trouble downloading HuggingFace models, -_^ this might help!!
+
+        For Linux:
+        export HF_ENDPOINT=https://hf-mirror.com
+
+        For Windows:
+        Good luck
+        ^_-
+
+        """
         self.ocr = OCR()
         if hasattr(self, "model_speciess"):
             self.layouter = LayoutRecognizer("layout." + self.model_speciess)
@@ -65,17 +83,6 @@ class RAGFlowPdfParser:
                 model_dir, "updown_concat_xgb.model"))
 
         self.page_from = 0
-        """
-        If you have trouble downloading HuggingFace models, -_^ this might help!!
-
-        For Linux:
-        export HF_ENDPOINT=https://hf-mirror.com
-
-        For Windows:
-        Good luck
-        ^_-
-
-        """
 
     def __char_width(self, c):
         return (c["x1"] - c["x0"]) // max(len(c["text"]), 1)
@@ -277,7 +284,11 @@ class RAGFlowPdfParser:
                 b["SP"] = ii
 
     def __ocr(self, pagenum, img, chars, ZM=3):
+        start = timer()
         bxs = self.ocr.detect(np.array(img))
+        logging.info(f"__ocr detecting boxes of a image cost ({timer() - start}s)")
+
+        start = timer()
         if not bxs:
             self.boxes.append([])
             return
@@ -308,14 +319,22 @@ class RAGFlowPdfParser:
             else:
                 bxs[ii]["text"] += c["text"]
 
+        logging.info(f"__ocr sorting {len(chars)} chars cost {timer() - start}s")
+        start = timer()
+        boxes_to_reg = []
+        img_np = np.array(img)
         for b in bxs:
             if not b["text"]:
                 left, right, top, bott = b["x0"] * ZM, b["x1"] * \
                                          ZM, b["top"] * ZM, b["bottom"] * ZM
-                b["text"] = self.ocr.recognize(np.array(img),
-                                               np.array([[left, top], [right, top], [right, bott], [left, bott]],
-                                                        dtype=np.float32))
+                b["box_image"] = self.ocr.get_rotate_crop_image(img_np, np.array([[left, top], [right, top], [right, bott], [left, bott]], dtype=np.float32))
+                boxes_to_reg.append(b)
             del b["txt"]
+        texts = self.ocr.recognize_batch([b["box_image"] for b in boxes_to_reg])
+        for i in range(len(boxes_to_reg)):
+            boxes_to_reg[i]["text"] = texts[i]
+            del boxes_to_reg[i]["box_image"]
+        logging.info(f"__ocr recognize {len(bxs)} boxes cost {timer() - start}s")
         bxs = [b for b in bxs if b["text"]]
         if self.mean_height[-1] == 0:
             self.mean_height[-1] = np.median([b["bottom"] - b["top"]
@@ -935,9 +954,12 @@ class RAGFlowPdfParser:
     @staticmethod
     def total_page_number(fnm, binary=None):
         try:
-            pdf = pdfplumber.open(
-                fnm) if not binary else pdfplumber.open(BytesIO(binary))
-            return len(pdf.pages)
+            with sys.modules[LOCK_KEY_pdfplumber]:
+                pdf = pdfplumber.open(
+                    fnm) if not binary else pdfplumber.open(BytesIO(binary))
+            total_page = len(pdf.pages)
+            pdf.close()
+            return total_page 
         except Exception:
             logging.exception("total_page_number")
 
@@ -951,20 +973,23 @@ class RAGFlowPdfParser:
         self.page_cum_height = [0]
         self.page_layout = []
         self.page_from = page_from
+        start = timer()
         try:
-            self.pdf = pdfplumber.open(fnm) if isinstance(
-                fnm, str) else pdfplumber.open(BytesIO(fnm))
-            self.page_images = [p.to_image(resolution=72 * zoomin).annotated for i, p in
-                                enumerate(self.pdf.pages[page_from:page_to])]
-            try:
-                self.page_chars = [[{**c, 'top': c['top'], 'bottom': c['bottom']} for c in page.dedupe_chars().chars if self._has_color(c)] for page in self.pdf.pages[page_from:page_to]]
-            except Exception as e:
-                logging.warning(f"Failed to extract characters for pages {page_from}-{page_to}: {str(e)}")
-                self.page_chars = [[] for _ in range(page_to - page_from)]  # If failed to extract, using empty list instead.
-                
-            self.total_page = len(self.pdf.pages)
+            with sys.modules[LOCK_KEY_pdfplumber]:
+                self.pdf = pdfplumber.open(fnm) if isinstance(
+                    fnm, str) else pdfplumber.open(BytesIO(fnm))
+                self.page_images = [p.to_image(resolution=72 * zoomin).annotated for i, p in
+                                    enumerate(self.pdf.pages[page_from:page_to])]
+                try:
+                    self.page_chars = [[c for c in page.dedupe_chars().chars if self._has_color(c)] for page in self.pdf.pages[page_from:page_to]]
+                except Exception as e:
+                    logging.warning(f"Failed to extract characters for pages {page_from}-{page_to}: {str(e)}")
+                    self.page_chars = [[] for _ in range(page_to - page_from)]  # If failed to extract, using empty list instead.
+                    
+                self.total_page = len(self.pdf.pages)
         except Exception:
             logging.exception("RAGFlowPdfParser __images__")
+        logging.info(f"__images__ dedupe_chars cost {timer() - start}s")
 
         self.outlines = []
         try:
@@ -981,8 +1006,11 @@ class RAGFlowPdfParser:
             dfs(outlines, 0)
         except Exception as e:
             logging.warning(f"Outlines exception: {e}")
+        finally:
+            self.pdf.close()
         if not self.outlines:
             logging.warning("Miss outlines")
+        
 
         logging.debug("Images converted.")
         self.is_english = [re.search(r"[a-zA-Z0-9,/¸;:'\[\]\(\)!@#$%^&*\"?<>._-]{30,}", "".join(
@@ -994,7 +1022,7 @@ class RAGFlowPdfParser:
         else:
             self.is_english = False
 
-        # st = timer()
+        start = timer()
         for i, img in enumerate(self.page_images):
             chars = self.page_chars[i] if not self.is_english else []
             self.mean_height.append(
@@ -1016,7 +1044,7 @@ class RAGFlowPdfParser:
             self.__ocr(i + 1, img, chars, zoomin)
             if callback and i % 6 == 5:
                 callback(prog=(i + 1) * 0.6 / len(self.page_images), msg="")
-        # print("OCR:", timer()-st)
+        logging.info(f"__images__ {len(self.page_images)} pages cost {timer() - start}s")
 
         if not self.is_english and not any(
                 [c for c in self.page_chars]) and self.boxes:
@@ -1142,7 +1170,7 @@ class RAGFlowPdfParser:
         return poss
 
 
-class PlainParser(object):
+class PlainParser:
     def __call__(self, filename, from_page=0, to_page=100000, **kwargs):
         self.outlines = []
         lines = []

deepdoc/parser/ppt_parser.py

@@ -19,7 +19,7 @@ from io import BytesIO
 from pptx import Presentation
 
 
-class RAGFlowPptParser(object):
+class RAGFlowPptParser:
     def __init__(self):
         super().__init__()
 

deepdoc/parser/resume/entities/corporations.py

@@ -30,10 +30,10 @@ GOODS = pd.read_csv(
 GOODS["cid"] = GOODS["cid"].astype(str)
 GOODS = GOODS.set_index(["cid"])
 CORP_TKS = json.load(
-    open(os.path.join(current_file_path, "res/corp.tks.freq.json"), "r")
+    open(os.path.join(current_file_path, "res/corp.tks.freq.json"), "r",encoding="utf-8")
 )
-GOOD_CORP = json.load(open(os.path.join(current_file_path, "res/good_corp.json"), "r"))
-CORP_TAG = json.load(open(os.path.join(current_file_path, "res/corp_tag.json"), "r"))
+GOOD_CORP = json.load(open(os.path.join(current_file_path, "res/good_corp.json"), "r",encoding="utf-8"))
+CORP_TAG = json.load(open(os.path.join(current_file_path, "res/corp_tag.json"), "r",encoding="utf-8"))
 
 
 def baike(cid, default_v=0):

deepdoc/parser/resume/entities/schools.py

@@ -25,7 +25,7 @@ TBL = pd.read_csv(
     os.path.join(current_file_path, "res/schools.csv"), sep="\t", header=0
 ).fillna("")
 TBL["name_en"] = TBL["name_en"].map(lambda x: x.lower().strip())
-GOOD_SCH = json.load(open(os.path.join(current_file_path, "res/good_sch.json"), "r"))
+GOOD_SCH = json.load(open(os.path.join(current_file_path, "res/good_sch.json"), "r",encoding="utf-8"))
 GOOD_SCH = set([re.sub(r"[,. &（）()]+", "", c) for c in GOOD_SCH])
 
 

deepdoc/parser/txt_parser.py

@@ -31,6 +31,7 @@ class RAGFlowTxtParser:
             raise TypeError("txt type should be str!")
         cks = [""]
         tk_nums = [0]
+        delimiter = delimiter.encode('utf-8').decode('unicode_escape').encode('latin1').decode('utf-8')
 
         def add_chunk(t):
             nonlocal cks, tk_nums, delimiter
@@ -51,11 +52,13 @@ class RAGFlowTxtParser:
             s = t
         if s < len(delimiter):
             dels.extend(list(delimiter[s:]))
-        dels = [re.escape(d) for d in delimiter if d]
+        dels = [re.escape(d) for d in dels if d]
         dels = [d for d in dels if d]
         dels = "|".join(dels)
         secs = re.split(r"(%s)" % dels, txt)
         for sec in secs:
+            if re.match(f"^{dels}$", sec):
+                continue
             add_chunk(sec)
 
         return [[c, ""] for c in cks]

deepdoc/vision/__init__.py

@@ -14,7 +14,8 @@
 #  limitations under the License.
 #
 import io
-
+import sys
+import threading
 import pdfplumber
 
 from .ocr import OCR
@@ -23,6 +24,11 @@ from .layout_recognizer import LayoutRecognizer4YOLOv10 as LayoutRecognizer
 from .table_structure_recognizer import TableStructureRecognizer
 
 
+LOCK_KEY_pdfplumber = "global_shared_lock_pdfplumber"
+if LOCK_KEY_pdfplumber not in sys.modules:
+    sys.modules[LOCK_KEY_pdfplumber] = threading.Lock()
+
+
 def init_in_out(args):
     from PIL import Image
     import os
@@ -36,12 +42,14 @@ def init_in_out(args):
 
     def pdf_pages(fnm, zoomin=3):
         nonlocal outputs, images
-        pdf = pdfplumber.open(fnm)
-        images = [p.to_image(resolution=72 * zoomin).annotated for i, p in
-                            enumerate(pdf.pages)]
+        with sys.modules[LOCK_KEY_pdfplumber]:
+            pdf = pdfplumber.open(fnm)
+            images = [p.to_image(resolution=72 * zoomin).annotated for i, p in
+                                enumerate(pdf.pages)]
 
         for i, page in enumerate(images):
             outputs.append(os.path.split(fnm)[-1] + f"_{i}.jpg")
+        pdf.close()
 
     def images_and_outputs(fnm):
         nonlocal outputs, images

deepdoc/vision/ocr.py

@@ -31,6 +31,7 @@ import onnxruntime as ort
 
 from .postprocess import build_post_process
 
+loaded_models = {}
 
 def transform(data, ops=None):
     """ transform """
@@ -67,6 +68,12 @@ def create_operators(op_param_list, global_config=None):
 
 def load_model(model_dir, nm):
     model_file_path = os.path.join(model_dir, nm + ".onnx")
+    global loaded_models
+    loaded_model = loaded_models.get(model_file_path)
+    if loaded_model:
+        logging.info(f"load_model {model_file_path} reuses cached model")
+        return loaded_model
+
     if not os.path.exists(model_file_path):
         raise ValueError("not find model file path {}".format(
             model_file_path))
@@ -102,18 +109,20 @@ def load_model(model_dir, nm):
             provider_options=[cuda_provider_options]
             )
         run_options.add_run_config_entry("memory.enable_memory_arena_shrinkage", "gpu:0")
-        logging.info(f"TextRecognizer {nm} uses GPU")
+        logging.info(f"load_model {model_file_path} uses GPU")
     else:
         sess = ort.InferenceSession(
             model_file_path,
             options=options,
             providers=['CPUExecutionProvider'])
         run_options.add_run_config_entry("memory.enable_memory_arena_shrinkage", "cpu")
-        logging.info(f"TextRecognizer {nm} uses CPU")
-    return sess, sess.get_inputs()[0], run_options
+        logging.info(f"load_model {model_file_path} uses CPU")
+    loaded_model = (sess, run_options)
+    loaded_models[model_file_path] = loaded_model
+    return loaded_model
 
 
-class TextRecognizer(object):
+class TextRecognizer:
     def __init__(self, model_dir):
         self.rec_image_shape = [int(v) for v in "3, 48, 320".split(",")]
         self.rec_batch_num = 16
@@ -123,7 +132,8 @@ class TextRecognizer(object):
             "use_space_char": True
         }
         self.postprocess_op = build_post_process(postprocess_params)
-        self.predictor, self.input_tensor, self.run_options = load_model(model_dir, 'rec')
+        self.predictor, self.run_options = load_model(model_dir, 'rec')
+        self.input_tensor = self.predictor.get_inputs()[0]
 
     def resize_norm_img(self, img, max_wh_ratio):
         imgC, imgH, imgW = self.rec_image_shape
@@ -383,7 +393,7 @@ class TextRecognizer(object):
         return rec_res, time.time() - st
 
 
-class TextDetector(object):
+class TextDetector:
     def __init__(self, model_dir):
         pre_process_list = [{
             'DetResizeForTest': {
@@ -408,7 +418,8 @@ class TextDetector(object):
                               "unclip_ratio": 1.5, "use_dilation": False, "score_mode": "fast", "box_type": "quad"}
 
         self.postprocess_op = build_post_process(postprocess_params)
-        self.predictor, self.input_tensor, self.run_options = load_model(model_dir, 'det')
+        self.predictor, self.run_options = load_model(model_dir, 'det')
+        self.input_tensor = self.predictor.get_inputs()[0]
 
         img_h, img_w = self.input_tensor.shape[2:]
         if isinstance(img_h, str) or isinstance(img_w, str):
@@ -495,7 +506,7 @@ class TextDetector(object):
         return dt_boxes, time.time() - st
 
 
-class OCR(object):
+class OCR:
     def __init__(self, model_dir=None):
         """
         If you have trouble downloading HuggingFace models, -_^ this might help!!
@@ -609,6 +620,16 @@ class OCR(object):
             return ""
         return text
 
+    def recognize_batch(self, img_list):
+        rec_res, elapse = self.text_recognizer(img_list)
+        texts = []
+        for i in range(len(rec_res)):
+            text, score = rec_res[i]
+            if score < self.drop_score:
+                text = ""
+            texts.append(text)
+        return texts
+
     def __call__(self, img, cls=True):
         time_dict = {'det': 0, 'rec': 0, 'cls': 0, 'all': 0}
 

deepdoc/vision/operators.py

@@ -23,7 +23,7 @@ import math
 from PIL import Image
 
 
-class DecodeImage(object):
+class DecodeImage:
     """ decode image """
 
     def __init__(self,
@@ -65,7 +65,7 @@ class DecodeImage(object):
         return data
 
 
-class StandardizeImage(object):
+class StandardizeImag:
     """normalize image
     Args:
         mean (list): im - mean
@@ -102,7 +102,7 @@ class StandardizeImage(object):
         return im, im_info
 
 
-class NormalizeImage(object):
+class NormalizeImage:
     """ normalize image such as subtract mean, divide std
     """
 
@@ -129,7 +129,7 @@ class NormalizeImage(object):
         return data
 
 
-class ToCHWImage(object):
+class ToCHWImage:
     """ convert hwc image to chw image
     """
 
@@ -145,19 +145,7 @@ class ToCHWImage(object):
         return data
 
 
-class Fasttext(object):
-    def __init__(self, path="None", **kwargs):
-        import fasttext
-        self.fast_model = fasttext.load_model(path)
-
-    def __call__(self, data):
-        label = data['label']
-        fast_label = self.fast_model[label]
-        data['fast_label'] = fast_label
-        return data
-
-
-class KeepKeys(object):
+class KeepKeys:
     def __init__(self, keep_keys, **kwargs):
         self.keep_keys = keep_keys
 
@@ -168,7 +156,7 @@ class KeepKeys(object):
         return data_list
 
 
-class Pad(object):
+class Pad:
     def __init__(self, size=None, size_div=32, **kwargs):
         if size is not None and not isinstance(size, (int, list, tuple)):
             raise TypeError("Type of target_size is invalid. Now is {}".format(
@@ -206,7 +194,7 @@ class Pad(object):
         return data
 
 
-class LinearResize(object):
+class LinearResize:
     """resize image by target_size and max_size
     Args:
         target_size (int): the target size of image
@@ -273,7 +261,7 @@ class LinearResize(object):
         return im_scale_y, im_scale_x
 
 
-class Resize(object):
+class Resize:
     def __init__(self, size=(640, 640), **kwargs):
         self.size = size
 
@@ -303,7 +291,7 @@ class Resize(object):
         return data
 
 
-class DetResizeForTest(object):
+class DetResizeForTest:
     def __init__(self, **kwargs):
         super(DetResizeForTest, self).__init__()
         self.resize_type = 0
@@ -433,7 +421,7 @@ class DetResizeForTest(object):
         return img, [ratio_h, ratio_w]
 
 
-class E2EResizeForTest(object):
+class E2EResizeForTest:
     def __init__(self, **kwargs):
         super(E2EResizeForTest, self).__init__()
         self.max_side_len = kwargs['max_side_len']
@@ -501,7 +489,7 @@ class E2EResizeForTest(object):
         return im, (ratio_h, ratio_w)
 
 
-class KieResize(object):
+class KieResize:
     def __init__(self, **kwargs):
         super(KieResize, self).__init__()
         self.max_side, self.min_side = kwargs['img_scale'][0], kwargs[
@@ -551,7 +539,7 @@ class KieResize(object):
         return points
 
 
-class SRResize(object):
+class SRResize:
     def __init__(self,
                  imgH=32,
                  imgW=128,
@@ -588,7 +576,7 @@ class SRResize(object):
         return data
 
 
-class ResizeNormalize(object):
+class ResizeNormalize:
     def __init__(self, size, interpolation=Image.BICUBIC):
         self.size = size
         self.interpolation = interpolation
@@ -600,7 +588,7 @@ class ResizeNormalize(object):
         return img_numpy
 
 
-class GrayImageChannelFormat(object):
+class GrayImageChannelFormat:
     """
     format gray scale image's channel: (3,h,w) -> (1,h,w)
     Args:
@@ -624,7 +612,7 @@ class GrayImageChannelFormat(object):
         return data
 
 
-class Permute(object):
+class Permute:
     """permute image
     Args:
         to_bgr (bool): whether convert RGB to BGR
@@ -647,7 +635,7 @@ class Permute(object):
         return im, im_info
 
 
-class PadStride(object):
+class PadStride:
     """ padding image for model with FPN, instead PadBatch(pad_to_stride) in original config
     Args:
         stride (bool): model with FPN need image shape % stride == 0

deepdoc/vision/postprocess.py

@@ -38,7 +38,7 @@ def build_post_process(config, global_config=None):
     return module_class(**config)
 
 
-class DBPostProcess(object):
+class DBPostProcess:
     """
     The post process for Differentiable Binarization (DB).
     """
@@ -259,7 +259,7 @@ class DBPostProcess(object):
         return boxes_batch
 
 
-class BaseRecLabelDecode(object):
+class BaseRecLabelDecode:
     """ Convert between text-label and text-index """
 
     def __init__(self, character_dict_path=None, use_space_char=False):

deepdoc/vision/recognizer.py

@@ -19,18 +19,16 @@ import os
 import math
 import numpy as np
 import cv2
-from copy import deepcopy
+from functools import cmp_to_key
 
-import onnxruntime as ort
-from huggingface_hub import snapshot_download
 
 from api.utils.file_utils import get_project_base_directory
 from .operators import *  # noqa: F403
 from .operators import preprocess
 from . import operators
+from .ocr import load_model
 
-
-class Recognizer(object):
+class Recognizer:
     def __init__(self, label_list, task_name, model_dir=None):
         """
         If you have trouble downloading HuggingFace models, -_^ this might help!!
@@ -47,51 +45,7 @@ class Recognizer(object):
             model_dir = os.path.join(
                         get_project_base_directory(),
                         "rag/res/deepdoc")
-            model_file_path = os.path.join(model_dir, task_name + ".onnx")
-            if not os.path.exists(model_file_path):
-                model_dir = snapshot_download(repo_id="InfiniFlow/deepdoc",
-                                              local_dir=os.path.join(get_project_base_directory(), "rag/res/deepdoc"),
-                                              local_dir_use_symlinks=False)
-                model_file_path = os.path.join(model_dir, task_name + ".onnx")
-        else:
-            model_file_path = os.path.join(model_dir, task_name + ".onnx")
-
-        if not os.path.exists(model_file_path):
-            raise ValueError("not find model file path {}".format(
-                model_file_path))
-
-        def cuda_is_available():
-            try:
-                import torch
-                if torch.cuda.is_available():
-                    return True
-            except Exception:
-                return False
-            return False
-
-        # https://github.com/microsoft/onnxruntime/issues/9509#issuecomment-951546580
-        # Shrink GPU memory after execution
-        self.run_options = ort.RunOptions()
-
-        if cuda_is_available():
-            options = ort.SessionOptions()
-            options.enable_cpu_mem_arena = False
-            cuda_provider_options = {
-                "device_id": 0, # Use specific GPU
-                "gpu_mem_limit": 512 * 1024 * 1024, # Limit gpu memory
-                "arena_extend_strategy": "kNextPowerOfTwo",  # gpu memory allocation strategy
-            }
-            self.ort_sess = ort.InferenceSession(
-                model_file_path, options=options,
-                providers=['CUDAExecutionProvider'],
-                provider_options=[cuda_provider_options]
-            )
-            self.run_options.add_run_config_entry("memory.enable_memory_arena_shrinkage", "gpu:0")
-            logging.info(f"Recognizer {task_name} uses GPU")
-        else:
-            self.ort_sess = ort.InferenceSession(model_file_path, providers=['CPUExecutionProvider'])
-            self.run_options.add_run_config_entry("memory.enable_memory_arena_shrinkage", "cpu")
-            logging.info(f"Recognizer {task_name} uses CPU")
+        self.ort_sess, self.run_options = load_model(model_dir, task_name)
         self.input_names = [node.name for node in self.ort_sess.get_inputs()]
         self.output_names = [node.name for node in self.ort_sess.get_outputs()]
         self.input_shape = self.ort_sess.get_inputs()[0].shape[2:4]
@@ -99,30 +53,22 @@ class Recognizer(object):
 
     @staticmethod
     def sort_Y_firstly(arr, threashold):
-        # sort using y1 first and then x1
-        arr = sorted(arr, key=lambda r: (r["top"], r["x0"]))
-        for i in range(len(arr) - 1):
-            for j in range(i, -1, -1):
-                # restore the order using th
-                if abs(arr[j + 1]["top"] - arr[j]["top"]) < threashold \
-                        and arr[j + 1]["x0"] < arr[j]["x0"]:
-                    tmp = deepcopy(arr[j])
-                    arr[j] = deepcopy(arr[j + 1])
-                    arr[j + 1] = deepcopy(tmp)
+        def cmp(c1, c2):
+            diff = c1["top"] - c2["top"]
+            if abs(diff) < threashold:
+                diff = c1["x0"] - c2["x0"]
+            return diff
+        arr = sorted(arr, key=cmp_to_key(cmp))
         return arr
 
     @staticmethod
-    def sort_X_firstly(arr, threashold, copy=True):
-        # sort using y1 first and then x1
-        arr = sorted(arr, key=lambda r: (r["x0"], r["top"]))
-        for i in range(len(arr) - 1):
-            for j in range(i, -1, -1):
-                # restore the order using th
-                if abs(arr[j + 1]["x0"] - arr[j]["x0"]) < threashold \
-                        and arr[j + 1]["top"] < arr[j]["top"]:
-                    tmp = deepcopy(arr[j]) if copy else arr[j]
-                    arr[j] = deepcopy(arr[j + 1]) if copy else arr[j + 1]
-                    arr[j + 1] = deepcopy(tmp) if copy else tmp
+    def sort_X_firstly(arr, threashold):
+        def cmp(c1, c2):
+            diff = c1["x0"] - c2["x0"]
+            if abs(diff) < threashold:
+                diff = c1["top"] - c2["top"]
+            return diff
+        arr = sorted(arr, key=cmp_to_key(cmp))
         return arr
 
     @staticmethod
@@ -145,8 +91,6 @@ class Recognizer(object):
                     arr[j + 1] = tmp
         return arr
 
-        return sorted(arr, key=lambda r: (r.get("C", r["x0"]), r["top"]))
-
     @staticmethod
     def sort_R_firstly(arr, thr=0):
         # sort using y1 first and then x1

deepdoc/vision/table_structure_recognizer.py

@@ -177,7 +177,7 @@ class TableStructureRecognizer(Recognizer):
         colwm = np.min(colwm) if colwm else 0
         crosspage = len(set([b["page_number"] for b in boxes])) > 1
         if crosspage:
-            boxes = Recognizer.sort_X_firstly(boxes, colwm / 2, False)
+            boxes = Recognizer.sort_X_firstly(boxes, colwm / 2)
         else:
             boxes = Recognizer.sort_C_firstly(boxes, colwm / 2)
         boxes[0]["cn"] = 0

docker/.env

@@ -80,13 +80,13 @@ REDIS_PASSWORD=infini_rag_flow
 SVR_HTTP_PORT=9380
 
 # The RAGFlow Docker image to download.
-# Defaults to the v0.16.0-slim edition, which is the RAGFlow Docker image without embedding models.
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0-slim
+# Defaults to the v0.17.2-slim edition, which is the RAGFlow Docker image without embedding models.
+RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2-slim
 #
 # To download the RAGFlow Docker image with embedding models, uncomment the following line instead:
-# RAGFLOW_IMAGE=infiniflow/ragflow:v0.16.0
+# RAGFLOW_IMAGE=infiniflow/ragflow:v0.17.2
 # 
-# The Docker image of the v0.16.0 edition includes:
+# The Docker image of the v0.17.2 edition includes:
 # - Built-in embedding models:
 #   - BAAI/bge-large-zh-v1.5
 #   - BAAI/bge-reranker-v2-m3
@@ -122,7 +122,7 @@ TIMEZONE='Asia/Shanghai'
 # HF_ENDPOINT=https://hf-mirror.com
 
 # Optimizations for MacOS
-# Uncomment the following line if your OS is MacOS:
+# Uncomment the following line if your operating system is MacOS:
 # MACOS=1
 
 # The maximum file size for each uploaded file, in bytes.
@@ -138,3 +138,11 @@ TIMEZONE='Asia/Shanghai'
 # - `ERROR`
 # For example, following line changes the log level of `ragflow.es_conn` to `DEBUG`:
 # LOG_LEVELS=ragflow.es_conn=DEBUG
+
+# aliyun OSS configuration
+# STORAGE_IMPL=OSS
+# ACCESS_KEY=xxx
+# SECRET_KEY=eee
+# ENDPOINT=http://oss-cn-hangzhou.aliyuncs.com
+# REGION=cn-hangzhou
+# BUCKET=ragflow65536

docker/README.md

@@ -78,8 +78,8 @@ The [.env](./.env) file contains important environment variables for Docker.
 - `RAGFLOW-IMAGE`  
   The Docker image edition. Available editions:  
   
-  - `infiniflow/ragflow:v0.16.0-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.16.0`: The RAGFlow Docker image with embedding models including:
+  - `infiniflow/ragflow:v0.17.2-slim` (default): The RAGFlow Docker image without embedding models.  
+  - `infiniflow/ragflow:v0.17.2`: The RAGFlow Docker image with embedding models including:
     - Built-in embedding models:
       - `BAAI/bge-large-zh-v1.5` 
       - `BAAI/bge-reranker-v2-m3`

docker/docker-compose-base.yml

@@ -3,7 +3,7 @@ services:
     container_name: ragflow-es-01
     profiles:
       - elasticsearch
-    image: docker.elastic.co/elasticsearch/elasticsearch:${STACK_VERSION}
+    image: elasticsearch:${STACK_VERSION}
     volumes:
       - esdata01:/usr/share/elasticsearch/data
     ports:
@@ -114,6 +114,7 @@ services:
     restart: on-failure
 
   redis:
+    # swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/valkey/valkey:8
     image: valkey/valkey:8
     container_name: ragflow-redis
     command: redis-server --requirepass ${REDIS_PASSWORD} --maxmemory 128mb --maxmemory-policy allkeys-lru

docker/docker-compose-gpu.yml

@@ -1,5 +1,5 @@
 # The RAGFlow team do not actively maintain docker-compose-gpu.yml, so use them at your own risk. 
-# However, you are welcome to file a pull request to improve it.
+# Pull requests to improve it are welcome.
 include:
   - ./docker-compose-base.yml
 

docker/entrypoint-parser.sh

@@ -15,8 +15,9 @@ CONSUMER_NO_BEG=$1
 CONSUMER_NO_END=$2
 
 function task_exe(){
+    JEMALLOC_PATH=$(pkg-config --variable=libdir jemalloc)/libjemalloc.so
     while [ 1 -eq 1 ]; do
-      $PY rag/svr/task_executor.py $1;
+      LD_PRELOAD=$JEMALLOC_PATH $PY rag/svr/task_executor.py $1;
     done
 }
 

docker/entrypoint.sh

@@ -17,8 +17,9 @@ if [[ -z "$WS" || $WS -lt 1 ]]; then
 fi
 
 function task_exe(){
+    JEMALLOC_PATH=$(pkg-config --variable=libdir jemalloc)/libjemalloc.so
     while [ 1 -eq 1 ];do
-      $PY rag/svr/task_executor.py $1;
+      LD_PRELOAD=$JEMALLOC_PATH $PY rag/svr/task_executor.py $1;
     done
 }
 

docker/launch_backend_service.sh

@@ -5,8 +5,10 @@ set -e
 
 # Unset HTTP proxies that might be set by Docker daemon
 export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
+export PYTHONPATH=$(pwd)
 
 export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu/
+JEMALLOC_PATH=$(pkg-config --variable=libdir jemalloc)/libjemalloc.so
 
 PY=python3
 
@@ -47,7 +49,7 @@ task_exe(){
     local retry_count=0
     while ! $STOP && [ $retry_count -lt $MAX_RETRIES ]; do
         echo "Starting task_executor.py for task $task_id (Attempt $((retry_count+1)))"
-        $PY rag/svr/task_executor.py "$task_id"
+        LD_PRELOAD=$JEMALLOC_PATH $PY rag/svr/task_executor.py "$task_id"
         EXIT_CODE=$?
         if [ $EXIT_CODE -eq 0 ]; then
             echo "task_executor.py for task $task_id exited successfully."
@@ -100,4 +102,4 @@ run_server &
 PIDS+=($!)
 
 # Wait for all background processes to finish
-wait
+wait

docker/service_conf.yaml.template

@@ -37,6 +37,13 @@ redis:
 #   access_key: 'access_key'
 #   secret_key: 'secret_key'
 #   region: 'region'
+# oss:
+#   access_key: '${ACCESS_KEY}'
+#   secret_key: '${SECRET_KEY}'
+#   endpoint_url: '${ENDPOINT}'
+#   region: '${REGION}'
+#   bucket: '${BUCKET}'
+#   prefix_path: '${OSS_PREFIX_PATH}'
 # azure:
 #   auth_type: 'sas'
 #   container_url: 'container_url'
@@ -52,6 +59,12 @@ redis:
 #   factory: 'Tongyi-Qianwen'
 #   api_key: 'sk-xxxxxxxxxxxxx'
 #   base_url: ''
+#   default_models:
+#     chat_model: 'qwen-plus'
+#     embedding_model: 'BAAI/bge-large-zh-v1.5@BAAI'
+#     rerank_model: ''
+#     asr_model: ''
+#     image2text_model: ''
 # oauth:
 #   github:
 #     client_id: xxxxxxxxxxxxxxxxxxxxxxxxx

docs/configurations.md

@@ -3,7 +3,7 @@ sidebar_position: 1
 slug: /configurations
 ---
 
-# Configurations
+# Configuration
 
 Configurations for deploying RAGFlow via Docker.
 
@@ -15,7 +15,7 @@ When it comes to system configurations, you will need to manage the following fi
 - [service_conf.yaml.template](https://github.com/infiniflow/ragflow/blob/main/docker/service_conf.yaml.template): Configures the back-end services. It specifies the system-level configuration for RAGFlow and is used by its API server and task executor. Upon container startup, the `service_conf.yaml` file will be generated based on this template file. This process replaces any environment variables within the template, allowing for dynamic configuration tailored to the container's environment.
 - [docker-compose.yml](https://github.com/infiniflow/ragflow/blob/main/docker/docker-compose.yml): The Docker Compose file for starting up the RAGFlow service.
 
-To update the default HTTP serving port (80), go to [docker-compose.yml](./docker/docker-compose.yml) and change `80:80`
+To update the default HTTP serving port (80), go to [docker-compose.yml](https://github.com/infiniflow/ragflow/blob/main/docker/docker-compose.yml) and change `80:80`
 to `<YOUR_SERVING_PORT>:80`.
 
 :::tip NOTE
@@ -97,8 +97,8 @@ The [.env](https://github.com/infiniflow/ragflow/blob/main/docker/.env) file con
 - `RAGFLOW-IMAGE`  
   The Docker image edition. Available editions:  
   
-  - `infiniflow/ragflow:v0.16.0-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.16.0`: The RAGFlow Docker image with embedding models including:
+  - `infiniflow/ragflow:v0.17.2-slim` (default): The RAGFlow Docker image without embedding models.  
+  - `infiniflow/ragflow:v0.17.2`: The RAGFlow Docker image with embedding models including:
     - Built-in embedding models:
       - `BAAI/bge-large-zh-v1.5` 
       - `BAAI/bge-reranker-v2-m3`

docs/develop/_category_.json

@@ -0,0 +1,8 @@
+{
+  "label": "Developers",
+  "position": 4,
+  "link": {
+    "type": "generated-index",
+    "description": "Guides for hardcore developers"
+  }
+}

docs/develop/acquire_ragflow_api_key.md

@@ -3,7 +3,7 @@ sidebar_position: 3
 slug: /acquire_ragflow_api_key
 ---
 
-# Acquire a RAGFlow API key
+# Acquire RAGFlow API key
 
 A key is required for the RAGFlow server to authenticate your requests via HTTP or a Python API. This documents provides instructions on obtaining a RAGFlow API key.
 
@@ -14,5 +14,5 @@ A key is required for the RAGFlow server to authenticate your requests via HTTP
 ![ragflow_api_key](https://github.com/user-attachments/assets/f461ed61-04c6-4faf-b3d8-6b5fa56be4e7)
 
 :::tip NOTE
-See the [RAGFlow HTTP API reference](../../references/http_api_reference.md) or the [RAGFlow Python API reference](../../references/python_api_reference.md) for a complete reference of RAGFlow's HTTP or Python APIs.
+See the [RAGFlow HTTP API reference](../references/http_api_reference.md) or the [RAGFlow Python API reference](../references/python_api_reference.md) for a complete reference of RAGFlow's HTTP or Python APIs.
 :::

docs/develop/build_docker_image.mdx

@@ -0,0 +1,92 @@
+---
+sidebar_position: 1
+slug: /build_docker_image
+---
+
+# Build RAGFlow Docker image
+import Tabs from '@theme/Tabs';
+import TabItem from '@theme/TabItem';
+
+A guide explaining how to build a RAGFlow Docker image from its source code. By following this guide, you'll be able to create a local Docker image that can be used for development, debugging, or testing purposes.
+
+## Target Audience
+
+- Developers who have added new features or modified the existing code and require a Docker image to view and debug their changes.
+- Developers seeking to build a RAGFlow Docker image for an ARM64 platform.
+- Testers aiming to explore the latest features of RAGFlow in a Docker image.
+
+## Prerequisites
+
+- CPU ≥ 4 cores
+- RAM ≥ 16 GB
+- Disk ≥ 50 GB
+- Docker ≥ 24.0.0 & Docker Compose ≥ v2.26.1
+
+## Build a Docker image
+
+<Tabs
+  defaultValue="without"
+  values={[
+    {label: 'Build a Docker image without embedding models', value: 'without'},
+    {label: 'Build a Docker image including embedding models', value: 'including'}
+  ]}>
+  <TabItem value="without">
+
+This image is approximately 2 GB in size and relies on external LLM and embedding services.
+
+:::danger IMPORTANT
+- While we also test RAGFlow on ARM64 platforms, we do not maintain RAGFlow Docker images for ARM. However, you can build an image yourself on a `linux/arm64` or `darwin/arm64` host machine as well.
+- For ARM64 platforms, please upgrade the `xgboost` version in **pyproject.toml** to `1.6.0` and ensure **unixODBC** is properly installed.
+:::
+
+```bash
+git clone https://github.com/infiniflow/ragflow.git
+cd ragflow/
+uv run download_deps.py
+docker build -f Dockerfile.deps -t infiniflow/ragflow_deps .
+docker build --build-arg LIGHTEN=1 -f Dockerfile -t infiniflow/ragflow:nightly-slim .
+```
+
+
+  </TabItem>
+  <TabItem value="including">
+
+This image is approximately 9 GB in size. As it includes embedding models, it relies on external LLM services only.  
+
+:::danger IMPORTANT
+- While we also test RAGFlow on ARM64 platforms, we do not maintain RAGFlow Docker images for ARM. However, you can build an image yourself on a `linux/arm64` or `darwin/arm64` host machine as well.
+- For ARM64 platforms, please upgrade the `xgboost` version in **pyproject.toml** to `1.6.0` and ensure **unixODBC** is properly installed.
+:::
+
+```bash
+git clone https://github.com/infiniflow/ragflow.git
+cd ragflow/
+uv run download_deps.py
+docker build -f Dockerfile.deps -t infiniflow/ragflow_deps .
+docker build -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
+  </TabItem>
+</Tabs>
+
+## Launch a RAGFlow Service from Docker for MacOS
+
+After building the infiniflow/ragflow:nightly-slim image, you are ready to launch a fully-functional RAGFlow service with all the required components, such as Elasticsearch, MySQL, MinIO, Redis, and more.
+
+## Example: Apple M2 Pro (Sequoia)
+
+1. Edit Docker Compose Configuration
+
+Open the `docker/.env` file. Find the `RAGFLOW_IMAGE` setting and change the image reference from `infiniflow/ragflow:v0.17.2-slim` to `infiniflow/ragflow:nightly-slim` to use the pre-built image.
+
+
+2. Launch the Service
+
+```bash
+cd docker
+$ docker compose -f docker-compose-macos.yml up -d
+```
+
+3. Access the RAGFlow Service
+
+Once the setup is complete, open your web browser and navigate to http://127.0.0.1 or your server's \<IP_ADDRESS\>; (the default port is \<PORT\> = 80). You will be directed to the RAGFlow welcome page. Enjoy!🍻

docs/develop/launch_ragflow_from_source.md

@@ -3,11 +3,11 @@ sidebar_position: 2
 slug: /launch_ragflow_from_source
 ---
 
-# Launch the RAGFlow Service from Source 
+# Launch service from source
 
 A guide explaining how to set up a RAGFlow service from its source code. By following this guide, you'll be able to debug using the source code.
 
-## Target Audience
+## Target audience
 
 Developers who have added new features or modified existing code and wish to debug using the source code, *provided that* their machine has the target deployment environment set up.
 
@@ -22,11 +22,11 @@ Developers who have added new features or modified existing code and wish to deb
 If you have not installed Docker on your local machine (Windows, Mac, or Linux), see the [Install Docker Engine](https://docs.docker.com/engine/install/) guide.
 :::
 
-## Launch the Service from Source
+## Launch a service from source
 
-To launch the RAGFlow service from source code:
+To launch a RAGFlow service from source code:
 
-### Clone the RAGFlow Repository
+### Clone the RAGFlow repository
 
 ```bash
 git clone https://github.com/infiniflow/ragflow.git
@@ -52,7 +52,7 @@ cd ragflow/
    ```
    *A virtual environment named `.venv` is created, and all Python dependencies are installed into the new environment.*
 
-### Launch Third-party Services
+### Launch third-party services
 
 The following command launches the 'base' services (MinIO, Elasticsearch, Redis, and MySQL) using Docker Compose:
 
@@ -70,7 +70,7 @@ docker compose -f docker/docker-compose-base.yml up -d
 
 2. In **docker/service_conf.yaml.template**, update mysql port to `5455` and es port to `1200`, as specified in **docker/.env**.
 
-### Launch the RAGFlow Backend Service
+### Launch the RAGFlow backend service
 
 1. Comment out the `nginx` line in **docker/entrypoint.sh**.
 

docs/faq.md

@@ -3,9 +3,9 @@ sidebar_position: 10
 slug: /faq
 ---
 
-# Frequently asked questions
+# FAQs
 
-Queries regarding general features, troubleshooting, usage, and more.
+Answers to questions about general features, troubleshooting, usage, and more.
 
 ---
 
@@ -22,6 +22,35 @@ The "garbage in garbage out" status quo remains unchanged despite the fact that
 
 ---
 
+### Where to find the version of RAGFlow? How to interpret it?
+
+You can find the RAGFlow version number on the **System** page of the UI:
+
+![Image](https://github.com/user-attachments/assets/20cf7213-2537-4e18-a88c-4dadf6228c6b)
+
+If you build RAGFlow from source, the version number is also in the system log:
+
+```
+        ____   ___    ______ ______ __               
+       / __ \ /   |  / ____// ____// /____  _      __
+      / /_/ // /| | / / __ / /_   / // __ \| | /| / /
+     / _, _// ___ |/ /_/ // __/  / // /_/ /| |/ |/ / 
+    /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/                             
+
+2025-02-18 10:10:43,835 INFO     1445658 RAGFlow version: v0.15.0-50-g6daae7f2 full
+```
+
+Where:
+
+- `v0.15.0`: The officially published release.
+- `50`: The number of git commits since the official release.
+- `g6daae7f2`: `g` is the prefix, and `6daae7f2` is the first seven characters of the current commit ID.
+- `full`/`slim`: The RAGFlow edition.
+  - `full`: The full RAGFlow edition.
+  - `slim`: The RAGFlow edition without embedding models and Python packages.
+
+---
+
 ### Why does it take longer for RAGFlow to parse a document than LangChain?
 
 We put painstaking effort into document pre-processing tasks like layout analysis, table structure recognition, and OCR (Optical Character Recognition) using our vision models. This contributes to the additional time required.
@@ -36,16 +65,16 @@ RAGFlow has a number of built-in models for document structure parsing, which ac
 
 ### Which architectures or devices does RAGFlow support?
 
-We officially support x86 CPU and nvidia GPU. While we also test RAGFlow on ARM64 platforms, we do not maintain RAGFlow Docker images for ARM. If you are on an ARM platform, follow [this guide](https://ragflow.io/docs/dev/build_docker_image) to build a RAGFlow Docker image.
+We officially support x86 CPU and nvidia GPU. While we also test RAGFlow on ARM64 platforms, we do not maintain RAGFlow Docker images for ARM. If you are on an ARM platform, follow [this guide](./develop/build_docker_image.mdx) to build a RAGFlow Docker image.
 
 ---
 
 ### Which embedding models can be deployed locally?
 
-RAGFlow offers two Docker image editions, `v0.16.0-slim` and `v0.16.0`:  
+RAGFlow offers two Docker image editions, `v0.17.2-slim` and `v0.17.2`:  
   
-- `infiniflow/ragflow:v0.16.0-slim` (default): The RAGFlow Docker image without embedding models.  
-- `infiniflow/ragflow:v0.16.0`: The RAGFlow Docker image with embedding models including:
+- `infiniflow/ragflow:v0.17.2-slim` (default): The RAGFlow Docker image without embedding models.  
+- `infiniflow/ragflow:v0.17.2`: The RAGFlow Docker image with embedding models including:
   - Built-in embedding models:
     - `BAAI/bge-large-zh-v1.5`
     - `BAAI/bge-reranker-v2-m3`
@@ -65,7 +94,7 @@ RAGFlow offers two Docker image editions, `v0.16.0-slim` and `v0.16.0`:
 
 ### Do you offer an API for integration with third-party applications?
 
-The corresponding APIs are now available. See the [RAGFlow HTTP API Reference](./http_api_reference.md) or the [RAGFlow Python API Reference](./python_api_reference.md) for more information.
+The corresponding APIs are now available. See the [RAGFlow HTTP API Reference](./references/http_api_reference.md) or the [RAGFlow Python API Reference](./references/python_api_reference.md) for more information.
 
 ---
 
@@ -87,7 +116,7 @@ Yes, we support enhancing user queries based on existing context of an ongoing c
 
 1. On the **Chat** page, hover over the desired assistant and select **Edit**.
 2. In the **Chat Configuration** popup, click the **Prompt Engine** tab.
-3. Toggle on **Multi-turn optimization** to enable this feature.
+3. Switch on **Multi-turn optimization** to enable this feature.
 
 ---
 
@@ -101,7 +130,7 @@ Yes, we support enhancing user queries based on existing context of an ongoing c
 
 #### How to build the RAGFlow image from scratch?
 
-See [Build a RAGFlow Docker image](https://ragflow.io/docs/dev/build_docker_image).
+See [Build a RAGFlow Docker image](./develop/build_docker_image.mdx).
 
 ---
 
@@ -267,7 +296,7 @@ tail -f ragflow/docker/ragflow-logs/*.log
    cd29bcb254bc   quay.io/minio/minio:RELEASE.2023-12-20T01-00-02Z       "/usr/bin/docker-ent…"   2 weeks ago    Up 11 hours      0.0.0.0:9001->9001/tcp, :::9001->9001/tcp, 0.0.0.0:9000->9000/tcp, :::9000->9000/tcp     ragflow-minio
    ```
 
-2. Follow [this document](../guides/run_health_check.md) to check the health status of the Elasticsearch service.
+2. Follow [this document](./guides/run_health_check.md) to check the health status of the Elasticsearch service.
 
 :::danger IMPORTANT
 The status of a Docker container status does not necessarily reflect the status of the service. You may find that your services are unhealthy even when the corresponding Docker containers are up running. Possible reasons for this include network failures, incorrect port numbers, or DNS issues.
@@ -289,7 +318,7 @@ The status of a Docker container status does not necessarily reflect the status
    91220e3285dd   docker.elastic.co/elasticsearch/elasticsearch:8.11.3   "/bin/tini -- /usr/l…"   11 hours ago   Up 11 hours (healthy)     9300/tcp, 0.0.0.0:9200->9200/tcp, :::9200->9200/tcp           ragflow-es-01
    ```
 
-2. Follow [this document](../guides/run_health_check.md) to check the health status of the Elasticsearch service.
+2. Follow [this document](./guides/run_health_check.md) to check the health status of the Elasticsearch service.
 
 :::danger IMPORTANT
 The status of a Docker container status does not necessarily reflect the status of the service. You may find that your services are unhealthy even when the corresponding Docker containers are up running. Possible reasons for this include network failures, incorrect port numbers, or DNS issues.
@@ -316,13 +345,13 @@ Your IP address or port number may be incorrect. If you are using the default co
 A correct Ollama IP address and port is crucial to adding models to Ollama:
 
 - If you are on demo.ragflow.io, ensure that the server hosting Ollama has a publicly accessible IP address. Note that 127.0.0.1 is not a publicly accessible IP address.
-- If you deploy RAGFlow locally, ensure that Ollama and RAGFlow are in the same LAN and can comunicate with each other.
+- If you deploy RAGFlow locally, ensure that Ollama and RAGFlow are in the same LAN and can communicate with each other.
 
-See [Deploy a local LLM](../guides/deploy_local_llm.mdx) for more information.
+See [Deploy a local LLM](./guides/models/deploy_local_llm.mdx) for more information.
 
 ---
 
-#### Do you offer examples of using deepdoc to parse PDF or other files?
+#### Do you offer examples of using DeepDoc to parse PDF or other files?
 
 Yes, we do. See the Python files under the **rag/app** folder.
 
@@ -366,7 +395,7 @@ Ensure that you update the **MAX_CONTENT_LENGTH** environment variable:
    cd29bcb254bc   quay.io/minio/minio:RELEASE.2023-12-20T01-00-02Z       "/usr/bin/docker-ent…"   2 weeks ago    Up 11 hours      0.0.0.0:9001->9001/tcp, :::9001->9001/tcp, 0.0.0.0:9000->9000/tcp, :::9000->9000/tcp     ragflow-minio
    ```
 
-2. Follow [this document](../guides/run_health_check.md) to check the health status of the Elasticsearch service.
+2. Follow [this document](./guides/run_health_check.md) to check the health status of the Elasticsearch service.
 
 :::danger IMPORTANT
 The status of a Docker container status does not necessarily reflect the status of the service. You may find that your services are unhealthy even when the corresponding Docker containers are up running. Possible reasons for this include network failures, incorrect port numbers, or DNS issues.
@@ -388,7 +417,7 @@ The status of a Docker container status does not necessarily reflect the status
 
 ### How to run RAGFlow with a locally deployed LLM?
 
-You can use Ollama or Xinference to deploy local LLM. See [here](../guides/deploy_local_llm.mdx) for more information.
+You can use Ollama or Xinference to deploy local LLM. See [here](./guides/models/deploy_local_llm.mdx) for more information.
 
 ---
 
@@ -405,7 +434,7 @@ If your model is not currently supported but has APIs compatible with those of O
 - If RAGFlow is locally deployed, ensure that your RAGFlow and Ollama are in the same LAN.
 - If you are using our online demo, ensure that the IP address of your Ollama server is public and accessible.
 
-See [here](../guides/deploy_local_llm.mdx) for more information.
+See [here](./guides/models/deploy_local_llm.mdx) for more information.
 
 ---
 
@@ -424,12 +453,12 @@ This error occurs because there are too many chunks matching your search criteri
 
 ### How to get an API key for integration with third-party applications?
 
-See [Acquire a RAGFlow API key](../guides/develop/acquire_ragflow_api_key.md).
+See [Acquire a RAGFlow API key](./develop/acquire_ragflow_api_key.md).
 
 ---
 
 ### How to upgrade RAGFlow?
 
-See [Upgrade RAGFlow](../guides/upgrade_ragflow.mdx) for more information.
+See [Upgrade RAGFlow](./guides/upgrade_ragflow.mdx) for more information.
 
 ---