fix(ollama/completion): add 'max_completion_token' support for ollama

litellm/llms/ollama/completion/transformation.py

@@ -150,6 +150,7 @@ class OllamaConfig(BaseConfig):
             "frequency_penalty",
             "stop",
             "response_format",
+            "max_completion_tokens",
         ]
 
     def map_openai_params(
@@ -160,7 +161,7 @@ class OllamaConfig(BaseConfig):
         drop_params: bool,
     ) -> dict:
         for param, value in non_default_params.items():
-            if param == "max_tokens":
+            if param == "max_tokens" or param == "max_completion_tokens":
                 optional_params["num_predict"] = value
             if param == "stream":
                 optional_params["stream"] = value
@@ -257,9 +258,13 @@ class OllamaConfig(BaseConfig):
         model_response.choices[0].finish_reason = "stop"
         if request_data.get("format", "") == "json":
             response_content = json.loads(response_json["response"])
-            
+
             # Check if this is a function call format with name/arguments structure
-            if isinstance(response_content, dict) and "name" in response_content and "arguments" in response_content:
+            if (
+                isinstance(response_content, dict)
+                and "name" in response_content
+                and "arguments" in response_content
+            ):
                 # Handle as function call (original behavior)
                 function_call = response_content
                 message = litellm.Message(

litellm/proxy/_new_secret_config.yaml

@@ -60,16 +60,9 @@ model_list:
   - model_name: gemini/gemini-2.0-flash
     litellm_params:
       model: gemini/gemini-2.0-flash
-
-litellm_settings:
-  num_retries: 0
-  check_provider_endpoint: true
-  cache: true
-  callbacks: ["otel"]
-
-files_settings:
-  - custom_llm_provider: gemini
-    api_key: os.environ/GEMINI_API_KEY
-
-general_settings:
-  store_prompts_in_spend_logs: true
+  - model_name: llama-qwen
+    litellm_params:
+      model: ollama/qwen2:0.5b
+    model_info:
+      input_cost_per_token: 0.75
+      output_cost_per_token: 3