Conversation
These three used to just break: use_bias/qk_norm changed which weights exist so loading crashed, and frozen running stats was flat-out rejected. Now mlx matches torch on all three, checked against real weight transplants (~1e-6) and a real checkpoint config shape.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to #509. I flagged three more spots where mlx and torch disagree on config handling, but left them out of scope since they just fail.
use_bias=Trueandqk_norm="none"change which weights exist (extra bias terms, missing RMSNorm params), so a checkpoint using either one crashed onload_safetensorswith a key mismatch.use_frozen_running_stats=Truewas rejected outright,from_hf_configraisedNotImplementedError.All three now work the same way in both backends.
use_biaswires into everyLinearinResidualBlockand the attention/FFN layers.qk_norm="none"skips building the query/key RMSNorm submodules entirely, same as torch. Frozen running stats clamps the RevIN mean/std forward from the context boundary instead of letting it keep updating into the horizon.Checked against real torch weights (
~1e-6match) and against a realTimesFM3Torch.to_dict(), to make surefrom_hf_configstill parses the production checkpoint shape correctly with and without these fields set.