Support 8 bit weights "unpacked" compute mode in MatmulNBits kernel #24959
Conversation
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
hariharans29
changed the title
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…hari/matmulnbits_8_bit_fallback
…microsoft/onnxruntime into hari/matmulnbits_8_bit_fallback
edgchen1
reviewed
Jun 6, 2025
edgchen1
reviewed
Jun 11, 2025
edgchen1
previously approved these changes
Jun 11, 2025
PR feedback Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com>
edgchen1
approved these changes
Jun 12, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Description
MLAS doesn't have optimized kernel implementations for 8-bit weights to support all user requested configurations (as allowed per the
MatMulNBitsop kernel spec) For example, onlyaccuracy_levelof 4 is supported when weights are 8 bits. In this case, provide a "fallback" execution path like the one available for 4 bit weights.This change:
Adds an execution path in the "unpacked compute" mode for 8-bit weights in
MatmulNBitskernelas MLASDequantize already supports 8 bit weights
1.1 Hitting the non-MLAS de-quantization path which doesn't support 8 bits should be very rare and an enforce is added to guard against hitting that
Added logging to alert users that the non-optimal "unpacked compute" mode is being used in the
MatmulNBitskernel when MLAS doesn't have an optimized kernel for that configurationMotivation and Context
Prevent users from hitting errors if MLAS doesn't host optimized kernels for 8 bit weights