Support Xinference (#320)

### What problem does this PR solve?

Issue link:#299

### Type of change

- [x] New Feature (non-breaking change which adds functionality)

rag/llm/__init__.py

@@ -21,6 +21,7 @@ from .cv_model import *
 EmbeddingModel = {
     "Ollama": OllamaEmbed,
     "OpenAI": OpenAIEmbed,
+    "Xinference": XinferenceEmbed,
     "Tongyi-Qianwen": HuEmbedding, #QWenEmbed,
     "ZHIPU-AI": ZhipuEmbed,
     "Moonshot": HuEmbedding
@@ -30,6 +31,7 @@ EmbeddingModel = {
 CvModel = {
     "OpenAI": GptV4,
     "Ollama": OllamaCV,
+    "Xinference": XinferenceCV,
     "Tongyi-Qianwen": QWenCV,
     "ZHIPU-AI": Zhipu4V,
     "Moonshot": LocalCV
@@ -41,6 +43,7 @@ ChatModel = {
     "ZHIPU-AI": ZhipuChat,
     "Tongyi-Qianwen": QWenChat,
     "Ollama": OllamaChat,
+    "Xinference": XinferenceChat,
     "Moonshot": MoonshotChat
 }
 

rag/llm/chat_model.py

@@ -158,6 +158,28 @@ class OllamaChat(Base):
             return "**ERROR**: " + str(e), 0
 
 
+class XinferenceChat(Base):
+    def __init__(self, key=None, model_name="", base_url=""):
+        self.client = OpenAI(api_key="xxx", base_url=base_url)
+        self.model_name = model_name
+
+    def chat(self, system, history, gen_conf):
+        if system:
+            history.insert(0, {"role": "system", "content": system})
+        try:
+            response = self.client.chat.completions.create(
+                model=self.model_name,
+                messages=history,
+                **gen_conf)
+            ans = response.choices[0].message.content.strip()
+            if response.choices[0].finish_reason == "length":
+                ans += "...\nFor the content length reason, it stopped, continue?" if is_english(
+                    [ans]) else "······\n由于长度的原因，回答被截断了，要继续吗？"
+            return ans, response.usage.completion_tokens
+        except openai.APIError as e:
+            return "**ERROR**: " + str(e), 0
+
+
 class LocalLLM(Base):
     class RPCProxy:
         def __init__(self, host, port):

rag/llm/cv_model.py

@@ -161,6 +161,22 @@ class OllamaCV(Base):
         except Exception as e:
             return "**ERROR**: " + str(e), 0
 
+class XinferenceCV(Base):
+    def __init__(self, key, model_name="", lang="Chinese", base_url=""):
+        self.client = OpenAI(api_key=key, base_url=base_url)
+        self.model_name = model_name
+        self.lang = lang
+
+    def describe(self, image, max_tokens=300):
+        b64 = self.image2base64(image)
+
+        res = self.client.chat.completions.create(
+            model=self.model_name,
+            messages=self.prompt(b64),
+            max_tokens=max_tokens,
+        )
+        return res.choices[0].message.content.strip(), res.usage.total_tokens
+
 
 class LocalCV(Base):
     def __init__(self, key, model_name="glm-4v", lang="Chinese", **kwargs):

rag/llm/embedding_model.py

@@ -170,3 +170,20 @@ class OllamaEmbed(Base):
         res = self.client.embeddings(prompt=text,
                                             model=self.model_name)
         return np.array(res["embedding"]), 128
+
+
+class XinferenceEmbed(Base):
+    def __init__(self, key, model_name="", base_url=""):
+        self.client = OpenAI(api_key="xxx", base_url=base_url)
+        self.model_name = model_name
+
+    def encode(self, texts: list, batch_size=32):
+        res = self.client.embeddings.create(input=texts,
+                                            model=self.model_name)
+        return np.array([d.embedding for d in res.data]
+                        ), res.usage.total_tokens
+
+    def encode_queries(self, text):
+        res = self.client.embeddings.create(input=[text],
+                                            model=self.model_name)
+        return np.array(res.data[0].embedding), res.usage.total_tokens