fix(pa): select the arch-native fp8 dtype in the metadata grid tuner - #984
Open
zjin-lcf wants to merge 1 commit into
Open
Conversation
The tuner built its K/V tensors as e4m3fnuz unconditionally. gfx950 consumes OCP e4m3, and the exponent-bias difference scales the decode output by 2x, so the tuner benchmarked correctly but then failed its own correctness gate and could not produce grid entries on gfx950. Pick the dtype from the runtime arch, matching the existing convention in qk_norm_rope_quant.py.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 tuner built its K/V tensors as e4m3fnuz unconditionally. gfx950 consumes OCP e4m3, and the exponent-bias difference scales the decode output by 2x, so the tuner benchmarked correctly but then failed its own correctness gate and could not produce grid entries on gfx950. Pick the dtype from the runtime arch, matching the existing convention in qk_norm_rope_quant.py.