feat: collection's lightrag model can be configured on web

aperag/api/components/schemas/model.yaml

@@ -114,4 +114,19 @@ availableEmbeddingList:
       type: array
       items:
         $ref: '#/availableEmbedding'
-
+
+availableModel:
+  type: object
+  properties:
+    model_service_provider:
+      type: string
+    model_name:
+      type: string
+
+availableModelList:
+  type: object
+  properties:
+    items:
+      type: array
+      items:
+        $ref: '#/availableModel'

aperag/api/openapi.yaml

@@ -71,6 +71,8 @@ paths:
     $ref: './paths/models.yaml#/modelServiceProvider'
   /available_embeddings:
     $ref: './paths/models.yaml#/availableEmbeddings'
+  /available_models:
+    $ref: './paths/models.yaml#/availableModels'
 
 
   # config

aperag/api/paths/models.yaml

@@ -34,6 +34,24 @@ availableEmbeddings:
             schema:
               $ref: '../components/schemas/common.yaml#/failResponse'
 
+availableModels:
+  get:
+    summary: Get available models
+    description: Get available models 
+    responses:
+      '200':
+        description: Available models
+        content:
+          application/json:
+            schema:
+              $ref: '../components/schemas/model.yaml#/availableModelList'
+      '401':
+        description: Unauthorized
+        content:
+          application/json:
+            schema:
+              $ref: '../components/schemas/common.yaml#/failResponse'
+
 modelServiceProvider:
   put:
     summary: Update model service provider

aperag/embed/base_embedding.py

@@ -28,7 +28,6 @@ def wrapper(*args, **kwargs):
 
     return wrapper
 
-
 _dimension_cache: dict[tuple[str, str], int] = {}
 
 

aperag/embed/embedding_service.py

@@ -57,9 +57,6 @@ def embed_query(self, text: str) -> List[float]:
         reraise=True,
     )
     def _embed_batch(self, batch: Sequence[str]) -> List[List[float]]:
-        """
-        单个 batch 的真正请求；加 tenacity 保证稳一点。
-        """
         response = litellm.embedding(
             model=self.model,
             api_base=self.api_base,

aperag/graph/lightrag_holder.py

@@ -2,6 +2,7 @@
 import logging
 from typing import Optional, List, Dict, Callable, Awaitable, Tuple, AsyncIterator, Any
 
+import json
 import numpy
 from lightrag import LightRAG, QueryParam
 from lightrag.kg.shared_storage import initialize_pipeline_status
@@ -10,6 +11,9 @@
 from lightrag.base import DocStatus
 
 from aperag.db.models import Collection
+from aperag.db.ops import (
+    query_msp_dict,
+)
 from aperag.embed.base_embedding import get_collection_embedding_model
 from aperag.utils.utils import generate_lightrag_namespace_prefix
 from config.settings import (
@@ -72,24 +76,40 @@ async def adelete_by_doc_id(self, doc_id: str) -> None:
 
 
 # ---------- Default llm_func & embed_impl ---------- #
-async def _default_llm_func(
-    prompt: str,
-    system_prompt: Optional[str] = None,
-    history_messages: List = [],
-    **kwargs,
-) -> str:
-    merged_kwargs = {
-        "api_key": LLM_API_KEY,
-        "base_url": LLM_BASE_URL,
-        "model": LLM_MODEL,
-        **kwargs,
-    }
-    return await openai_complete_if_cache(
-        prompt=prompt,
-        system_prompt=system_prompt,
-        history_messages=history_messages,
-        **merged_kwargs,
-    )
+async def gen_lightrag_llm_func(collection: Collection) -> Callable[..., Awaitable[str]]:
+    config = json.loads(collection.config)
+    lightrag_backend = config.get("lightrag_model_service_provider", "")
+    lightrag_model_name = config.get("lightrag_model_name", "")
+    logging.info("gen_lightrag_llm_func %s %s", lightrag_backend, lightrag_model_name)
+
+    msp_dict = await query_msp_dict(collection.user)
+    if lightrag_backend in msp_dict:
+        msp = msp_dict[lightrag_backend]
+        lightrag_model_service_url = msp.base_url
+        lightrag_model_service_api_key = msp.api_key
+        logging.info("gen_lightrag_llm_func %s %s", lightrag_model_service_url, lightrag_model_service_api_key)
+
+        async def lightrag_llm_func(
+            prompt: str,
+            system_prompt: Optional[str] = None,
+            history_messages: List = [],
+            **kwargs,
+        ) -> str:
+            merged_kwargs = {
+                "api_key": lightrag_model_service_api_key,
+                "base_url": lightrag_model_service_url,
+                "model": lightrag_model_name,
+                **kwargs,
+            }
+            return await openai_complete_if_cache(
+                prompt=prompt,
+                system_prompt=system_prompt,
+                history_messages=history_messages,
+                **merged_kwargs,
+            )
+        return lightrag_llm_func
+
+    return None
 
 # Module-level cache
 _lightrag_instances: Dict[str, LightRagHolder] = {}
@@ -145,9 +165,9 @@ async def lightrag_embed_func(texts: list[str]) -> numpy.ndarray:
     return lightrag_embed_func, dim
 
 async def get_lightrag_holder(
-    collection: Collection,
-    llm_func: Callable[..., Awaitable[str]] = _default_llm_func,
+    collection: Collection
 ) -> LightRagHolder:
+    # Fixme: if lightrag_model changes, we need to re-initialize the lightrag instance
     namespace_prefix: str = generate_lightrag_namespace_prefix(collection.id)
     if not namespace_prefix or not isinstance(namespace_prefix, str):
         raise ValueError("A valid namespace_prefix string must be provided.")
@@ -162,6 +182,7 @@ async def get_lightrag_holder(
         logger.info(f"Initializing LightRAG instance for namespace '{namespace_prefix}' (lazy loading)...")
         try:
             embed_func, dim = await gen_lightrag_embed_func(collection=collection)
+            llm_func = await gen_lightrag_llm_func(collection=collection)
             client = await _create_and_initialize_lightrag(namespace_prefix, llm_func, embed_func, embed_dim=dim)
             _lightrag_instances[namespace_prefix] = client
             logger.info(f"LightRAG instance for namespace '{namespace_prefix}' initialized successfully.")

aperag/tasks/index.py

@@ -243,8 +243,8 @@ def add_index_for_document(self, document_id):
             }
             document.relate_ids = json.dumps(relate_ids)
 
-            enable_light_rag = config.get("enable_light_rag", True)
-            if enable_light_rag:
+            enable_lightrag = config.get("enable_lightrag", True)
+            if enable_lightrag:
                 add_lightrag_index(content, document, local_doc)
 
     except FeishuNoPermission:
@@ -355,8 +355,8 @@ def update_index_for_document(self, document_id):
         document.relate_ids = json.dumps(relate_ids)
         logger.info(f"update qdrant points: {document.relate_ids} for document {local_doc.path}")
 
-        enable_light_rag = config.get("enable_light_rag", True)
-        if enable_light_rag:
+        enable_lightrag = config.get("enable_lightrag", True)
+        if enable_lightrag:
             add_lightrag_index(content, document, local_doc)
 
     except FeishuNoPermission:

aperag/views/main.py

@@ -1157,6 +1157,25 @@ async def list_available_embeddings(request) -> view_models.AvailableEmbeddingLi
     return success(view_models.AvailableEmbeddingList( items=response,))
 
 
+
+@router.get("/available_models")
+async def list_available_models(request) -> view_models.AvailableModelList:
+    user = get_user(request)
+    supported_msp_dict = {supported_msp["name"]: supported_msp for supported_msp in settings.SUPPORTED_MODEL_SERVICE_PROVIDERS}
+    msp_list = await query_msp_list(user)
+    logger.info(msp_list)
+    response = []
+    for msp in msp_list:
+        if msp.name in supported_msp_dict:
+            supported_msp = supported_msp_dict[msp.name]
+            for model in supported_msp.get("models", []):
+                response.append(view_models.AvailableModel(
+                    model_service_provider=msp.name,
+                    model_name=model,
+                ))
+    return success(view_models.AvailableModelList( items=response,))
+
+
 def default_page(request, exception):
     return render(request, '404.html')
 

aperag/views/models.py

@@ -1,6 +1,6 @@
 # generated by datamodel-codegen:
 #   filename:  openapi.merged.yaml
-#   timestamp: 2025-04-26T07:25:07+00:00
+#   timestamp: 2025-04-27T06:17:52+00:00
 
 from __future__ import annotations
 
@@ -391,6 +391,15 @@ class AvailableEmbeddingList(BaseModel):
     items: Optional[list[AvailableEmbedding]] = None
 
 
+class AvailableModel(BaseModel):
+    model_service_provider: Optional[str] = None
+    model_name: Optional[str] = None
+
+
+class AvailableModelList(BaseModel):
+    items: Optional[list[AvailableModel]] = None
+
+
 class Auth0(BaseModel):
     auth_domain: Optional[str] = None
     auth_app_id: Optional[str] = None

config/settings.py

@@ -258,6 +258,18 @@
             "text-embedding-3-small",
             "text-embedding-3-large",
             "text-embedding-ada-002",
+        ],
+        "models": [
+            "gpt-3.5-turbo",
+            "gpt-4",
+            "gpt-4-turbo",
+            "gpt-4o-mini",
+            "gpt-4o",
+            "o1",
+            "o1-mini",
+            "o3",
+            "o3-mini",
+            "o4-mini",
         ]
     },
     {
@@ -269,13 +281,31 @@
             "text-embedding-v1",
             "text-embedding-v2",
             "text-embedding-v3",
+        ],
+        "models": [
+            "deepseek-r1",
+            "deepseek-v3",
+            "qwen-max",
+            "qwen-long",
+            "qwen-plus",
+            "qwen-plus-latest",
+            "qwen-turbo",
+            "qwq-32b",
+            "qwq-plus",
+            "qwq-plus-latest",
+            "qwen-vl-max",
+            "qwen-vl-plus"
         ]
     },
     {
         "name": "deepseek",
         "label": "DeepSeek",
         "allow_custom_base_url": False,
-        "base_url": "https://api.deepseek.com/v1"
+        "base_url": "https://api.deepseek.com/v1",
+        "models": [
+            "deepseek-r1",
+            "deepseek-v3"
+        ]
     },
     {
         "name": "siliconflow",
@@ -286,6 +316,11 @@
             "BAAI/bge-large-en-v1.5",
             "BAAI/bge-large-zh-v1.5",
             "BAAI/bge-m3",
+        ],
+        "models": [
+            "Qwen/QwQ-32B",
+            "deepseek-ai/Deepseek-R1",
+            "deepseek-ai/Deepseek-V3",
         ]
     },
 ]

envs/env.template

@@ -72,11 +72,7 @@ CHAT_CONSUMER_IMPLEMENTATION=document-qa
 # RETRIEVE_MODE is one of: classic, graph, mix
 RETRIEVE_MODE=classic
 
-# --- LLM Settings ---
-LIGHT_RAG_LLM_API_KEY=
-LIGHT_RAG_LLM_BASE_URL=
-LIGHT_RAG_LLM_MODEL=
-# --- General Settings ---
+# --- LIGHT RAG General Settings ---
 LIGHT_RAG_WORKING_DIR=
 LIGHT_RAG_ENABLE_LLM_CACHE=
 LIGHT_RAG_MAX_PARALLEL_INSERT=