Litellm本地模型调用配置问题，求助

[diedcow1]

openclaw.json中关于ollama本地模型直接调用（"baseUrl": "http://127.0.0.1:11434",）LiteLLM聚合的配置如下（"baseUrl": "http://127.0.0.1:4000/v1",），

直接指定ollama的qwen3.5:9b，在web界面及tg bot是可以对话的，但是模型切换到litellm 就报错（⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.），但终端执行命令curl http://127.0.0.1:4000/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer sk-litellm-zsy-2026" -d '{"model": "LiteLLM","messages": [{"role": "user","content": "介绍一下你自己"}]}' 就没问题，是哪里配置不对吗？：

"ollama": {
"baseUrl": "http://127.0.0.1:11434",
"apiKey": "sk-litellm-zsy-2026",
"api": "ollama",
"models": [
{
"id": "ollama/qwen3.5:9b",
"name": "qwen3.5:9b",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 262144,
"maxTokens": 8192
}
]
},
"litellm": {
"baseUrl": "http://127.0.0.1:4000/v1",
"apiKey": "sk-litellm-zsy-2026",
"api": "openai-completions",
"models": [
{
"id": "LiteLLM",
"name": "LiteLLM 聚合节点",
"api": "openai-completions",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 128000,
"maxTokens": 8192
}
]
}
 
LiteLLM的配置文件config.yaml写的是api_base: http://127.0.0.1:11434:
model_list:

• model_name: LiteLLM
  litellm_params:
  #model: ollama/qwen3.5:9b
  #model: ollama/deepseek-v3.1:671b-cloud
  #model: ollama/deepseek-r1:14b
  model: ollama/qwen3.5:9b
  api_base: http://127.0.0.1:11434
  model_info:
  id: ollama-local
  router_settings:
  routing_strategy: least-busy
  num_retries: 1
  retry_after: 5
  cooldown_time: 60
  litellm_settings:
  request_timeout: 900
  drop_params: true
  general_settings:
  master_key: "sk-litellm-zsy-2026"
  database_url: "postgresql://postgres:Lit22sdf23@localhost:5432/litellm_db"
  api

[smqd19]

This kind of error is usually a mismatch between the model name/config in Litellm and what’s actually available or exposed at your backend, especially when you’re aggregating via Litellm. A few things worth checking based on how we’ve deployed local Ollama models behind aggregators:

• Model ID/Name Consistency:
  Your curl command works because you’re directly hitting Litellm with the model name "LiteLLM". But in config, you’ve got "id": "LiteLLM" for Litellm and "id": "ollama/qwen3.5:9b" for Ollama. Aggregators like Litellm often need to route based on explicit model names. Make sure that the model parameter in your API call matches the name/id Litellm expects, which might be "qwen3.5:9b" or "ollama/qwen3.5:9b".

• Model Exposure:
  Double-check that Litellm is configured to expose the Ollama-backed model. In production, we’ve had to set allowed_models or similar in Litellm config to ensure it’s actually listing/accepting requests for that model.

• API Type:
  You’ve got "api": "openai-completions" for Litellm. If your Ollama backend is using a different protocol, Litellm might not route correctly. Try running /v1/models on Litellm and see if your model shows up.
  Example:

  curl http://127.0.0.1:4000/v1/models -H "Authorization: Bearer sk-litellm-zsy-2026"

• Debug Logs:
  Enable debug logging in Litellm and check exactly what model name it’s receiving and what it routes to. In our setups, mismatched names between config and request are almost always the culprit.

• Quick Test:
  Try POSTing with "model": "qwen3.5:9b" instead of "LiteLLM", or use the exact id from the config. That often solves these routing hiccups.

Let me know if you see the model in /v1/models output—if not, it’s a config issue; if yes, probably a request mismatch.