add supprot for OpenAI-API-Compatible llm (#1787)

### What problem does this PR solve?

#1771 add supprot for OpenAI-API-Compatible

### Type of change

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

---------

Co-authored-by: Zhedong Cen <[email protected]>

api/apps/llm_app.py

@@ -129,6 +129,9 @@ def add_llm():
     elif factory == "LocalAI":
         llm_name = req["llm_name"]+"___LocalAI"
         api_key = "xxxxxxxxxxxxxxx"
+    elif factory == "OpenAI-API-Compatible":
+        llm_name = req["llm_name"]+"___OpenAI-API"
+        api_key = req["api_key"]
     else:
         llm_name = req["llm_name"]
         api_key = "xxxxxxxxxxxxxxx"
@@ -145,7 +148,7 @@ def add_llm():
     msg = ""
     if llm["model_type"] == LLMType.EMBEDDING.value:
         mdl = EmbeddingModel[factory](
-            key=llm['api_key'] if factory in ["VolcEngine", "Bedrock"] else None,
+            key=llm['api_key'] if factory in ["VolcEngine", "Bedrock","OpenAI-API-Compatible"] else None,
             model_name=llm["llm_name"], 
             base_url=llm["api_base"])
         try:
@@ -156,7 +159,7 @@ def add_llm():
             msg += f"\nFail to access embedding model({llm['llm_name']})." + str(e)
     elif llm["model_type"] == LLMType.CHAT.value:
         mdl = ChatModel[factory](
-            key=llm['api_key'] if factory in ["VolcEngine", "Bedrock"] else None,
+            key=llm['api_key'] if factory in ["VolcEngine", "Bedrock","OpenAI-API-Compatible"] else None,
             model_name=llm["llm_name"],
             base_url=llm["api_base"]
         )
@@ -181,7 +184,7 @@ def add_llm():
                 e)
     elif llm["model_type"] == LLMType.IMAGE2TEXT.value:
         mdl = CvModel[factory](
-            key=None, model_name=llm["llm_name"], base_url=llm["api_base"]
+            key=llm["api_key"] if factory in ["OpenAI-API-Compatible"] else None, model_name=llm["llm_name"], base_url=llm["api_base"]
         )
         try:
             img_url = (

conf/llm_factories.json

@@ -158,6 +158,13 @@
             "status": "1",
             "llm": []
         },
+        {
+            "name": "OpenAI-API-Compatible",
+            "logo": "",
+            "tags": "LLM,TEXT EMBEDDING,SPEECH2TEXT,MODERATION",
+            "status": "1",
+            "llm": []
+        },
         {
             "name": "Moonshot",
             "logo": "",

rag/llm/__init__.py

@@ -36,7 +36,8 @@ EmbeddingModel = {
     "Bedrock": BedrockEmbed,
     "Gemini": GeminiEmbed,
     "NVIDIA": NvidiaEmbed,
-    "LM-Studio": LmStudioEmbed
+    "LM-Studio": LmStudioEmbed,
+    "OpenAI-API-Compatible": OpenAI_APIEmbed
 }
 
 
@@ -53,7 +54,8 @@ CvModel = {
     "LocalAI": LocalAICV,
     "NVIDIA": NvidiaCV,
     "LM-Studio": LmStudioCV,
-    "StepFun":StepFunCV
+    "StepFun":StepFunCV,
+    "OpenAI-API-Compatible": OpenAI_APICV
 }
 
 
@@ -78,7 +80,8 @@ ChatModel = {
     "OpenRouter": OpenRouterChat,
     "StepFun": StepFunChat,
     "NVIDIA": NvidiaChat,
-    "LM-Studio": LmStudioChat
+    "LM-Studio": LmStudioChat,
+    "OpenAI-API-Compatible": OpenAI_APIChat
 }
 
 
@@ -88,7 +91,8 @@ RerankModel = {
     "Youdao": YoudaoRerank,
     "Xinference": XInferenceRerank,
     "NVIDIA": NvidiaRerank,
-    "LM-Studio": LmStudioRerank
+    "LM-Studio": LmStudioRerank,
+    "OpenAI-API-Compatible": OpenAI_APIRerank
 }
 
 

rag/llm/chat_model.py

@@ -887,6 +887,16 @@ class LmStudioChat(Base):
         if not base_url:
             raise ValueError("Local llm url cannot be None")
         if base_url.split("/")[-1] != "v1":
-            self.base_url = os.path.join(base_url, "v1")
-        self.client = OpenAI(api_key="lm-studio", base_url=self.base_url)
+            base_url = os.path.join(base_url, "v1")
+        self.client = OpenAI(api_key="lm-studio", base_url=base_url)
         self.model_name = model_name
+
+
+class OpenAI_APIChat(Base):
+    def __init__(self, key, model_name, base_url):
+        if not base_url:
+            raise ValueError("url cannot be None")
+        if base_url.split("/")[-1] != "v1":
+            base_url = os.path.join(base_url, "v1")
+        model_name = model_name.split("___")[0]
+        super().__init__(key, model_name, base_url)

rag/llm/cv_model.py

@@ -638,3 +638,14 @@ class LmStudioCV(GptV4):
         self.client = OpenAI(api_key="lm-studio", base_url=base_url)
         self.model_name = model_name
         self.lang = lang
+
+
+class OpenAI_APICV(GptV4):
+    def __init__(self, key, model_name, base_url, lang="Chinese"):
+        if not base_url:
+            raise ValueError("url cannot be None")
+        if base_url.split("/")[-1] != "v1":
+            base_url = os.path.join(base_url, "v1")
+        self.client = OpenAI(api_key=key, base_url=base_url)
+        self.model_name = model_name.split("___")[0]
+        self.lang = lang

rag/llm/embedding_model.py

@@ -513,3 +513,13 @@ class LmStudioEmbed(LocalAIEmbed):
             self.base_url = os.path.join(base_url, "v1")
         self.client = OpenAI(api_key="lm-studio", base_url=self.base_url)
         self.model_name = model_name
+
+
+class OpenAI_APIEmbed(OpenAIEmbed):
+    def __init__(self, key, model_name, base_url):
+        if not base_url:
+            raise ValueError("url cannot be None")
+        if base_url.split("/")[-1] != "v1":
+            self.base_url = os.path.join(base_url, "v1")
+        self.client = OpenAI(api_key=key, base_url=base_url)
+        self.model_name = model_name.split("___")[0]

rag/llm/rerank_model.py

@@ -212,3 +212,11 @@ class LmStudioRerank(Base):
 
     def similarity(self, query: str, texts: list):
         raise NotImplementedError("The LmStudioRerank has not been implement")
+
+
+class OpenAI_APIRerank(Base):
+    def __init__(self, key, model_name, base_url):
+        pass
+
+    def similarity(self, query: str, texts: list):
+        raise NotImplementedError("The api has not been implement")

web/src/interfaces/request/llm.ts

@@ -3,6 +3,7 @@ export interface IAddLlmRequestBody {
   llm_name: string;
   model_type: string;
   api_base?: string; // chat|embedding|speech2text|image2text
+  api_key: string;
 }
 
 export interface IDeleteLlmRequestBody {

web/src/pages/user-setting/constants.tsx

@@ -17,4 +17,4 @@ export const UserSettingIconMap = {
 
 export * from '@/constants/setting';
 
-export const LocalLlmFactories = ['Ollama', 'Xinference','LocalAI','LM-Studio'];
+export const LocalLlmFactories = ['Ollama', 'Xinference','LocalAI','LM-Studio',"OpenAI-API-Compatible"];

web/src/pages/user-setting/setting-model/constant.ts

@@ -21,7 +21,8 @@ export const IconMap = {
   LocalAI: 'local-ai',
   StepFun: 'stepfun',
   NVIDIA:'nvidia',
-  'LM-Studio':'lm-studio'
+  'LM-Studio':'lm-studio',
+  'OpenAI-API-Compatible':'openai-api'
 };
 
 export const BedrockRegionList = [

web/src/pages/user-setting/setting-model/ollama-modal/index.tsx

@@ -92,6 +92,13 @@ const OllamaModal = ({
         >
           <Input placeholder={t('baseUrlNameMessage')} />
         </Form.Item>
+        <Form.Item<FieldType>
+          label={t('apiKey')}
+          name="api_key"
+          rules={[{ required: false, message: t('apiKeyMessage') }]}
+        >
+          <Input placeholder={t('apiKeyMessage')} />
+        </Form.Item>
         <Form.Item noStyle dependencies={['model_type']}>
           {({ getFieldValue }) =>
             getFieldValue('model_type') === 'chat' && (