Fix Flux quality: don't quantize text encoder#115
Draft
stikves wants to merge 2 commits into
Draft
Conversation
The Flux text encoder (Qwen-based, 3B) was marked quantizable=True, causing it to be exported at INT4 when --compression 4bit is used. This degrades prompt adherence (cosine 0.9655 vs reference) because the conditioning signal loses semantic precision at 4-bit. Set quantizable=False for the Flux text encoder so it stays at f16 regardless of global compression setting. The transformer (4B) remains INT4 — it's quality-insensitive since it just predicts noise. Fixes apple#111.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The Flux text encoder (Qwen-based, 3B) was marked quantizable=True, causing it to be exported at INT4 when --compression 4bit is used. This degrades prompt adherence (cosine 0.9655 vs reference) because the conditioning signal loses semantic precision at 4-bit.
Set quantizable=False for the Flux text encoder so it stays at f16 regardless of global compression setting. The transformer (4B) remains INT4 — it's quality-insensitive since it just predicts noise.
Fixes #111.