Repository navigation
Conversation
Replace the old thinking-mode defaults (temp 0.6, top_p 0.95, rep_penalty 1.05) with the official Qwen3.5-122B-A10B Instruct/non-thinking profile: temp 0.7, top_p 0.8, top_k 20, min_p 0.0, repetition_penalty 1.0, presence_penalty 1.5. presence_penalty is the actual anti-repetition lever per the model card; repetition_penalty stays at 1.0. Adds the greedy-decode warning and the thinking-mode profile.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Updates the Recommended sampling section to the official Qwen3.5-122B-A10B Instruct / non-thinking profile.
Key change: presence_penalty (not repetition_penalty) is the model card's recommended anti-repetition lever, set between 0 and 2, higher risks language mixing. Also documents the never-greedy-decode warning and the thinking-mode profile. Source: the official Qwen/Qwen3.5-122B-A10B model card.