Reasoning models like QwQ or DeepSeek-R1 break chat title generation

[crockpotveggies]

What happened?

when the UI calls https://<server>/api/convos/gen_title it triggers a 404 and sends back the following:
message "Title not found or method not implemented for the conversation's endpoint"

strangely the LibreChat server logs don't show anything. I think this may be either a race condition or an incompatibility with how reasoning models work.

Alternatively in librechat.yaml using the titleModel param does not seem to fix this issue. Instead, the /gen_title endpoint returns a 200 OK and the title becomes New Chat.

In the debug logs, it looks like LibreChat is not calling the correct model:

2025-03-28T17:40:01.576Z debug: Making request to http://jgpt.local:8000/v1/chat/completions
2025-03-28T17:40:01.578Z warn: [OpenAIClient.chatCompletion][create] API error
2025-03-28T17:40:01.578Z error
2025-03-28T17:40:01.578Z error: [OpenAIClient.chatCompletion] Unhandled error type Error: 404 status code (no body)
2025-03-28T17:40:01.578Z error: [OpenAIClient] There was an issue generating the title with the completion method Error: 404 status code (no body)

Using a MITM proxy I can see it is requesting the model from the incorrect endpoint - instead of trying to call the model from the IPEX endpoint (see librechat.yaml config below) it calls the same model from the current vLLM endpoint that is not hosting the model. Is this a limitation or a bug?

Version Information

$ docker images | grep librechat
ghcr.io/danny-avila/librechat-dev latest e80f2070a5eb 9 days ago 906MB
ghcr.io/danny-avila/librechat-rag-api-dev-lite latest 6550e7ddf180 3 weeks ago 1.3GB

$ git rev-parse HEAD
d6a1778

Steps to Reproduce

1. Set up multiple vLLM endpoints, one using a reasoning model like QwQ-32B and another using a standard chat LLM like Qwen2.5-7B
2. Modify librechat.yaml file to use these custom endpoints. For the endpoint hosting reasoning model(s), specify the titleModel you want to use coming from a different custom endpoint.
3. Create a new chat and prompt "hi".
4. The UI will retain the chat title as "New Chat".

What browsers are you seeing the problem on?

Firefox

Relevant log output

2025-03-28T17:37:13.949Z warn: RAG API is either not running or not reachable at http://rag_api:8000, you may experience errors with file uploads.
2025-03-28T17:37:13.969Z info: [MCP] Initializing servers
2025-03-28T17:37:13.970Z info: [MCP][exa] Connection state: connecting
2025-03-28T17:37:13.973Z debug: [MCP][exa] Transport sending: {"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"librechat-mcp-cl... [truncated]
2025-03-28T17:37:14.306Z debug: [MCP][exa] Transport sending: {"method":"notifications/initialized","jsonrpc":"2.0"}
2025-03-28T17:37:14.306Z info: [MCP][exa] Connection state: connected
2025-03-28T17:37:14.306Z info: [MCP][exa] Capabilities: {"tools":{}}
2025-03-28T17:37:14.306Z debug: [MCP][exa] Transport sending: {"method":"tools/list","jsonrpc":"2.0","id":1}
2025-03-28T17:37:14.307Z info: [MCP][exa] Available tools: search
2025-03-28T17:37:14.308Z info: [MCP] Initialized 1/1 server(s)
2025-03-28T17:37:14.308Z info: [MCP][exa] ✓ Initialized
2025-03-28T17:37:14.308Z info: [MCP] All servers initialized successfully
2025-03-28T17:37:14.308Z debug: [MCP][exa] Transport sending: {"method":"tools/list","jsonrpc":"2.0","id":2}
2025-03-28T17:37:14.310Z info: No changes needed for 'USER' role permissions
2025-03-28T17:37:14.311Z info: No changes needed for 'ADMIN' role permissions
2025-03-28T17:37:14.311Z info: Outdated Config version: 1.1.4
Latest version: 1.2.3

      Check out the Config changelogs for the latest options and features added.

      https://... [truncated]
2025-03-28T17:37:14.314Z info: Server listening on all interfaces at port 80. Use http://localhost:80 to access it
2025-03-28T17:39:17.724Z debug: [validateJson] files
["\"Ai_PDF.json\"","\"BrowserOp.json\"","\"Dr_Thoths_Tarot.json\"","\"DreamInterpreter.json\"","\"VoxScript.json\"","\"askyourpdf.json\"","\"drink_maestro.json\"","\"earthImagesAndVisualizations.json\"","\"image_prompt_enhancer.json\"","\"qrCodes.json\"","\"scholarai.json\"","\"uberchord.json\"","\"web_search.json\""]
2025-03-28T17:39:54.319Z debug: [AskController]
{
  text: "hi",
  conversationId: null,
  endpoint: "vLLM CUDA",
  endpointType: "custom",
  resendFiles: true,
    modelOptions.model: "Qwen/QwQ-32B-AWQ",
  modelsConfig: "exists",
}
2025-03-28T17:39:54.391Z debug: [BaseClient] Loading history:
{
  conversationId: "c27b35f5-ad5c-499d-ae0d-66790db5c121",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
}
2025-03-28T17:39:54.430Z debug: [BaseClient] Context Count (1/2)
{
  remainingContextTokens: 4087,
  maxContextTokens: 4095,
}
2025-03-28T17:39:54.430Z debug: [BaseClient] Context Count (2/2)
{
  remainingContextTokens: 4087,
  maxContextTokens: 4095,
}
2025-03-28T17:39:54.430Z debug: [BaseClient] tokenCountMap:
{
  6323f556-0dcb-4fc0-9d85-810385e4c875: 5,
}
2025-03-28T17:39:54.431Z debug: [BaseClient]
{
  promptTokens: 8,
  remainingContextTokens: 4087,
  payloadSize: 1,
  maxContextTokens: 4095,
}
2025-03-28T17:39:54.431Z debug: [BaseClient] tokenCountMap
{
  6323f556-0dcb-4fc0-9d85-810385e4c875: 5,
  instructions: undefined,
}
2025-03-28T17:39:54.431Z debug: [BaseClient] userMessage
{
  messageId: "6323f556-0dcb-4fc0-9d85-810385e4c875",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
  conversationId: "c27b35f5-ad5c-499d-ae0d-66790db5c121",
  sender: "User",
  text: "hi",
  isCreatedByUser: true,
  tokenCount: 5,
}
2025-03-28T17:39:54.432Z debug: [OpenAIClient] chatCompletion
{
  baseURL: "http://jgpt.local:8000/v1",
    modelOptions.model: "Qwen/QwQ-32B-AWQ",
    modelOptions.user: "67d8962ce74996275c71b4e2",
    modelOptions.stream: true,
    // 1 message(s)
    modelOptions.messages: [{"role":"user","content":"hi"}],
}
2025-03-28T17:39:54.435Z debug: Making request to http://jgpt.local:8000/v1/chat/completions
2025-03-28T17:39:54.443Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-03-28T17:40:01.556Z debug: [OpenAIClient] chatCompletion response
{
  object: "chat.completion",
  id: "chatcmpl-7c122dcd444f4fde8ee78a2e2ca3074c",
  // 1 choice(s)
  choices: [{"stop_reason":null,"message":{"role":"assistant","reasoning_content":".\n","content":"\n\nHello! 😊... [truncated]],
  created: 1743183594,
  model: "Qwen/QwQ-32B-AWQ",
}
2025-03-28T17:40:01.558Z debug: [spendTokens] conversationId: c27b35f5-ad5c-499d-ae0d-66790db5c121 | Context: message | Token usage:
{
  promptTokens: 8,
  completionTokens: 243,
}
2025-03-28T17:40:01.564Z debug: [spendTokens] No transactions incurred against balance
2025-03-28T17:40:01.567Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-03-28T17:40:01.572Z debug: [AskController] Request closed
2025-03-28T17:40:01.575Z debug: [OpenAIClient] chatCompletion
{
  baseURL: "http://jgpt.local:8000/v1",
    modelOptions.model: "Qwen/Qwen2.5-7B-Instruct-AWQ",
    modelOptions.user: "67d8962ce74996275c71b4e2",
    modelOptions.temperature: 0.2,
    modelOptions.presence_penalty: 0,
    modelOptions.frequency_penalty: 0,
    modelOptions.max_tokens: 16,
    // 1 message(s)
    modelOptions.messages: [{"role":"system","content":"Please generate a concise, 5-word-or-less title for the conversation, us... [truncated]],
}
2025-03-28T17:40:01.576Z debug: Making request to http://jgpt.local:8000/v1/chat/completions
2025-03-28T17:40:01.578Z warn: [OpenAIClient.chatCompletion][create] API error
2025-03-28T17:40:01.578Z error
2025-03-28T17:40:01.578Z error: [OpenAIClient.chatCompletion] Unhandled error type Error: 404 status code (no body)
2025-03-28T17:40:01.578Z error: [OpenAIClient] There was an issue generating the title with the completion method Error: 404 status code (no body)
2025-03-28T17:40:01.578Z debug: [OpenAIClient] Convo Title: New Chat
2025-03-28T17:40:01.578Z debug: [saveConvo] api/server/services/Endpoints/openAI/addTitle.js

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Do those models accept system messages? that might be the issue you're having. It's not specific to reasoning models, as Deepseek-R1 titling works from other providers

[danny-avila]

Nevermind, I understand the situation better, you are expecting this which is not yet implemented:

#3321