[GH-ISSUE #1572] [FEAT]: Add ability to rewrite parameters in the json request. #1018

Open
opened 2026-02-22 18:22:42 -05:00 by yindo · 1 comment
Owner

Originally created by @goodglitch on GitHub (May 30, 2024).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/1572

What would you like to see?

In case when "max_tokens" field is none my LLM server updates the field with too small tokens number. More precisely it is around 500 and most my coding requests got truncated. Please, add a button in the Options/LLM preferences that for each LLM can add item to an update dict with parameter and its value, like "max_tokens: 2048 or "top_k: 3". Then update json request with this dict. On the top there may be button to reset (clear update dict).

This feature will allow to fix many personal user problems by themselves and I think will be very welcomed by the community. Meanwhile, I have to edit LLM server code each time I update it, since all the changes got erased.

Originally created by @goodglitch on GitHub (May 30, 2024). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/1572 ### What would you like to see? In case when "max_tokens" field is none my LLM server updates the field with too small tokens number. More precisely it is around 500 and most my coding requests got truncated. Please, add a button in the Options/LLM preferences that for each LLM can add item to an update dict with parameter and its value, like "max_tokens: 2048 or "top_k: 3". Then update json request with this dict. On the top there may be button to reset (clear update dict). This feature will allow to fix many personal user problems by themselves and I think will be very welcomed by the community. Meanwhile, I have to edit LLM server code each time I update it, since all the changes got erased.
yindo added the enhancementfeature request labels 2026-02-22 18:22:42 -05:00
yindo changed title from [FEAT]: Add ability to rewrite parameters in the json request. to [GH-ISSUE #1572] [FEAT]: Add ability to rewrite parameters in the json request. 2026-06-05 14:38:30 -04:00
Author
Owner

@ppil commented on GitHub (Mar 9, 2026):

Testing qwen3.5, this feature is more of a requirement to get that model functioning in an instruct mode. Even with reasoning-budget 0 and thinking disabled via template kwargs, the model disobeys and will still generate thought patterns unless the presence_penalty is set. I can set this via cli args on llama-server, but the API request that gets processed from anythingllm has presence_penalty set to 0, which takes precedence. With models becoming more dependent on these parameters, I think this should be revisited.

<!-- gh-comment-id:4028484153 --> @ppil commented on GitHub (Mar 9, 2026): Testing qwen3.5, this feature is more of a requirement to get that model functioning in an instruct mode. Even with reasoning-budget 0 and thinking disabled via template kwargs, the model disobeys and will still generate thought patterns unless the presence_penalty is set. I can set this via cli args on llama-server, but the API request that gets processed from anythingllm has presence_penalty set to 0, which takes precedence. With models becoming more dependent on these parameters, I think this should be revisited.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#1018