Serverless models in custom endpoint get chat title generation failure

[mansiibm]

What happened?

Hi using the version for LC-62b4f3b79576ed549bc6e586415952653d4c42a4 and the chat title is failing although in the backend the chat title completion is done.
The config-

endpoints:
custom:
- name: "Deepseek"
apiKey: "${DEEPSEEK_API_KEY}"
baseURL: "https://region.api.cognitive.microsoft.com/models"
models:
default: ["DeepSeek-R1"]
fetch: false
titleConvo: true
directEndpoint: false
titleModel: "DeepSeek-V3-0324"
modelDisplayLabel: "Deepseek"

both the models are deployed and active .

Version Information

62b4f3b LC version.

Steps to Reproduce

1. Deploy the 62b4f3b tag
2. Use deepseek
3. Model responding but chat title failing

What browsers are you seeing the problem on?

Microsoft Edge

Relevant log output

Chat title not generated for UI.

Screenshots

Code of Conduct

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

[danny-avila]

Reasoning models are not good for generating titles (there's a time limit, and reasoning tokens are not accounted for the title).

Set a different title model:

https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint#titlemodel

[mansiibm]

Hi Danny , thanks for replying, I used the same deepseek-r1 model for title generation still it failed and from azure I do not have other models apart from DeepSeek-V3-0324 for title generation, also this did not happen until this version - danny-avila/librechat-dev-api:dff4fcac00939cf5a9b2077eba4cadcfe47c72eb only from 7.9-rc1 this started happening.

[mansiibm]

Hi Danny with the advent of 7.9 , title is getting generated but using the model mentioned in the bracket only . For example if the following is my config-

endpoints:
custom:
- name: "Deepseek"
apiKey: "${DEEPSEEK_API_KEY}"
baseURL: "https://region.api.cognitive.microsoft.com/models"
models:
default: ["DeepSeek-R1"]
fetch: false
titleConvo: true
directEndpoint: true
titleModel: "DeepSeek-V3-0324"
modelDisplayLabel: "Deepseek"

so the chat title is there -

on interaction with the model but the V3 as it is not a thinking model- it is not being used, was working until - ghcr.io/danny-avila/librechat-dev-api:dff4fcac00939cf5a9b2077eba4cadcfe47c72eb with the same setting as above. But title generation occurs for it

I kept fetch to true also for 7.9 still no change.

so if I change the config to -

endpoints:
custom:
- name: "Deepseek"
apiKey: "${DEEPSEEK_API_KEY}"
baseURL: "https://region.api.cognitive.microsoft.com/models"
models:
default: ["DeepSeek-R1", "DeepSeek-V3-0324"]
fetch: false
titleConvo: true
directEndpoint: true
titleModel: "DeepSeek-V3-0324"
modelDisplayLabel: "Deepseek"

then the response is as -

So for the thinking model- r1 the title from R1 is getting generated hence the "think" but if I try chatting from the V3 model then the title generation is getting done as expected, but I do not need the V3 in the model list , need only in the title generation.

Please have a look at the title generation for this , thanks !!!

[mansiibm]

I even use the titleEndpoint-

still did not work , still the think and all-

[mansiibm]

Hi, any response?

[mansiibm]

Hi Danny, I am using the tag- 545a909 . So one thing I observed , that for custom endpoints with following config-

endpoints:
custom:
- name: "Deepseek"
apiKey: "${DEEPSEEK_API_KEY}"
baseURL: "https://some-region.api.cognitive.microsoft.com/models"
models:
default: ["DeepSeek-R1"]
fetch: false
titleConvo: true
directEndpoint: true
modelDisplayLabel: "Deepseek"
- name: "xai"
apiKey: "${XAI_API_KEY}"
baseURL: "https://some-region.api.cognitive.microsoft.com/models"
models:
default: ["grok-3", "grok-3-mini"]
fetch: false
titleConvo: true
directEndpoint: true
titleModel: "grok-3"
modelDisplayLabel: "xAI"

Now only for the models which are present in the default list of models will use themselves for generating the title . If suppose I try to chat with the grok-3-mini model then the chat title is not generated ,

similarly for deepseek , the deepseek -r1 is being use to generate the title no matter what title model is set

[mansiibm]

For this setup -

titleConvo
titleMethod
titleEndpoint
do not work

[mansiibm]

Only until the

models:
default:

The setup works but beyond that it does not recognize the config.

also earlier I could use the complete URL- https://region.api.cognitive.microsoft.com/models/chat/completions?api-version=2024-05-01-preview and set directEndpoint to true now it works only until -/models and is not affected by directEndpoint.