From 5c81e01de5bc0b10107bbff06d63c8dd971702d8 Mon Sep 17 00:00:00 2001
From: Yongteng Lei <yongtengrey@outlook.com>
Date: Wed, 3 Dec 2025 11:15:45 +0800
Subject: [PATCH] Fix: incorrect async chat streamly output (#11679)

### What problem does this PR solve?

Incorrect async chat streamly output. #11677.

Disable beartype for #11666.

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
---
 agent/__init__.py                  |  4 ++--
 api/__init__.py                    |  4 ++--
 api/db/services/llm_service.py     | 11 ++++++++++-
 rag/__init__.py                    |  4 ++--
 sdk/python/ragflow_sdk/__init__.py |  2 +-
 5 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/agent/__init__.py b/agent/__init__.py
index 643f79713..a42cd9a6d 100644
--- a/agent/__init__.py
+++ b/agent/__init__.py
@@ -14,5 +14,5 @@
 #  limitations under the License.
 #
 
-from beartype.claw import beartype_this_package
-beartype_this_package()
+# from beartype.claw import beartype_this_package
+# beartype_this_package()
diff --git a/api/__init__.py b/api/__init__.py
index 643f79713..a42cd9a6d 100644
--- a/api/__init__.py
+++ b/api/__init__.py
@@ -14,5 +14,5 @@
 #  limitations under the License.
 #
 
-from beartype.claw import beartype_this_package
-beartype_this_package()
+# from beartype.claw import beartype_this_package
+# beartype_this_package()
diff --git a/api/db/services/llm_service.py b/api/db/services/llm_service.py
index 0b23bfc72..6a63713ec 100644
--- a/api/db/services/llm_service.py
+++ b/api/db/services/llm_service.py
@@ -385,6 +385,7 @@ class LLMBundle(LLM4Tenant):
 
     async def async_chat_streamly(self, system: str, history: list, gen_conf: dict = {}, **kwargs):
         total_tokens = 0
+        ans = ""
         if self.is_tools and self.mdl.is_tools:
             stream_fn = getattr(self.mdl, "async_chat_streamly_with_tools", None)
         else:
@@ -397,7 +398,15 @@ class LLMBundle(LLM4Tenant):
                 if isinstance(txt, int):
                     total_tokens = txt
                     break
-                yield txt
+
+                if txt.endswith("</think>"):
+                    ans = ans[: -len("</think>")]
+
+                if not self.verbose_tool_use:
+                    txt = re.sub(r"<tool_call>.*?</tool_call>", "", txt, flags=re.DOTALL)
+
+                ans += txt
+                yield ans
             if total_tokens and not TenantLLMService.increase_usage(self.tenant_id, self.llm_type, total_tokens, self.llm_name):
                 logging.error("LLMBundle.async_chat_streamly can't update token usage for {}/CHAT llm_name: {}, used_tokens: {}".format(self.tenant_id, self.llm_name, total_tokens))
             return
diff --git a/rag/__init__.py b/rag/__init__.py
index 643f79713..a42cd9a6d 100644
--- a/rag/__init__.py
+++ b/rag/__init__.py
@@ -14,5 +14,5 @@
 #  limitations under the License.
 #
 
-from beartype.claw import beartype_this_package
-beartype_this_package()
+# from beartype.claw import beartype_this_package
+# beartype_this_package()
diff --git a/sdk/python/ragflow_sdk/__init__.py b/sdk/python/ragflow_sdk/__init__.py
index 7535aa4ed..ea383cfc3 100644
--- a/sdk/python/ragflow_sdk/__init__.py
+++ b/sdk/python/ragflow_sdk/__init__.py
@@ -37,4 +37,4 @@ __all__ = [
     "Document",
     "Chunk",
     "Agent"
-]
\ No newline at end of file
+]