Cannot start a chat #6424
-
What happened?Images attached. I am trying to use simple roleplay scenarios (short prompts) that the LLM backends have no problems doing on open-webui. How do I fix this? I legitimately want to use librechat, but these are some pretty strong barriers. 
Version Informationdocker images | grep librechat Steps to ReproducePrompt: https://gist.github.com/frenzybiscuit/3d864440089629cddca25b781b4037a4 What browsers are you seeing the problem on?Firefox Relevant log outputScreenshotsNo response Code of Conduct
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 7 replies
-
|
Changing the max tokens manually in the prompt template fixes this, but the end user isn't going to know they have to do that, most likely. So it's not really a fix. Is there a way to set this setting in the librechat.yaml? |
Beta Was this translation helpful? Give feedback.
-
|
The problem is that LC doesn't recognize the model you are using ("Fallen-*") so a low default context window is being used (4095). Use model specs if you want an easy setting for your users: You can customize a "preset" this way, so you can remove the "addParams" field from the endpoint/system level and instead use the modelSpec to define |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! I missed the maxContextTokens, which is exactly what I need. To confirm though, if I do this the user wont be able to modify it in the prompt settings? Is there a way to set the default context window higher in the source code, so it applies to every model without restricting the end user from changing the setting? |
Beta Was this translation helpful? Give feedback.
-
Correct, they wouldn't have to adjust parameters. They can still reproduce the error if they post something larger than the context window you set.
No it currently works by model name (including partial matches), something like this will be implemented, though, both customizing lengths for models and defining a default for all. A model spec will be good enough as it would limit your users to whatever options you set, anyway. |
Beta Was this translation helpful? Give feedback.
-
I will add that with model specs, the user can still adjust the parameters if you enable the interface elements that it disables. This is explained in the docs I added. |
Beta Was this translation helpful? Give feedback.
-
|
So I added this and it works (great) but I can no longer select the agents tab (bad). Click on agents doesn't do anything. Video: Screen.Recording.2025-03-19.at.3.22.04.PM.movCode: |
Beta Was this translation helpful? Give feedback.
-
https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/model_specs#overview You can enable everything from before # in your yaml
interface:
agents: true
prompts: true
parameters: true
presets: true
endpointsMenu: true
modelSelect: true
bookmarks: true
multiConvo: true
modelSpecs:
enforce: false
prioritize: false
list:
- name: "Azure OpenAI GPT-4"
# etcWe're going to improve the UI/UX when these settings are enabled together. |
Beta Was this translation helpful? Give feedback.
-
|
Okay, thanks for putting up with my BS as I get familiar with LibreChat. Appreciate you answering my dumb questions. |
Beta Was this translation helpful? Give feedback.
Correct, they wouldn't have to adjust parameters. They can still reproduce the error if they post something larger than the context window you set.
No it currently works by model name (including partial matches), something like this will be implemented, though, both customizing lengths for models and defining a default for all. A model spec will be good enough as it would limit your users to whatever options you set, anyway.