(feat) add mode for config.yaml health checks

litellm/health_check.py

@@ -50,7 +50,6 @@ async def _perform_health_check(model_list: list):
         try:
             await litellm.aembedding(**model_params)
         except Exception as e:
-            print_verbose(f"\n\n Got Exception, {e}")
             print_verbose(f"Health check failed for model {model_params['model']}. Error: {e}")
             return False
         return True
@@ -60,11 +59,6 @@ async def _perform_health_check(model_list: list):
         try:
             await litellm.acompletion(**model_params)
         except Exception as e:            
-            print_verbose(f"\n\n Got Exception, {e}")
-            error_str = (str(e))
-            if "This is not a chat model" in error_str or "The chatCompletion operation does not work with the specified model" in error_str:
-                    return await _check_embedding_model(model_params)
-                
             print_verbose(f"Health check failed for model {model_params['model']}. Error: {e}")
             return False
         
@@ -78,7 +72,8 @@ async def _perform_health_check(model_list: list):
         litellm_params["messages"] = _get_random_llm_message()
 
         prepped_params.append(litellm_params)
-        if "embedding" in litellm_params["model"]:
+        if model.get("mode", None) == "embedding":
+            # this is an embedding model
             tasks.append(_check_embedding_model(litellm_params))
         else:
             tasks.append(_check_model(litellm_params))

litellm/proxy/proxy_config.yaml

@@ -11,6 +11,7 @@ model_list:
       api_base: os.environ/AZURE_API_BASE
       api_key: os.environ/AZURE_API_KEY
       api_version: "2023-07-01-preview"
+    mode: embedding
   - model_name: openai-gpt-3.5
     litellm_params:
       model: gpt-3.5-turbo
@@ -19,6 +20,7 @@ model_list:
     litellm_params:
       model: text-embedding-ada-002
       api_key: os.environ/OPENAI_API_KEY
+    mode: embedding
 
 
 litellm_settings:

litellm/proxy/types.py

@@ -40,6 +40,7 @@ class ProxyChatCompletionRequest(BaseModel):
 
 class ModelInfo(BaseModel):
     id: Optional[str]
+    mode: Optional[str]
 
     class Config:
         extra = Extra.allow  # Allow extra fields