Add marlin int4 kernel #333
Merged
Conversation
This file contains 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
dacorvo
force-pushed
the
add_marlin_int4_kernel
branch
from
6803193
to
4d7bd36
Compare
Original fix in vLLM project: The reason for the crash was the inline PTX assembly that introduced the async_copy with streaming behavior. The solution is to use the more standard PTX for async_copy (without the fractional L2 policy for "evict_first"). There is no performance difference between standard async_copy PTX and the previous one.
This is to guarantee Marlin kernels output is similar to the output obtained using dequantized weights.
Adding more tests revealed a bug in the Marlin int4 kernel when the weights and inputs are large enough. Failing configurations are marked as xfail.
dacorvo
force-pushed
the
add_marlin_int4_kernel
branch
from
4d7bd36
to
ce90118
Compare
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Outdated suggestions cannot be applied.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.
What does this PR do?
This adds a modified Marlin fp16/int4 kernel to the library and creates two new QTensor subclasses to use it:
There are issues with the weight/scales/zero-point readback as soon as parallelization increases. The consequence is that output features higher than 128 are corrupted when a sufficient amount of inputs are parallelized.
As a consequence, the AWQ kernel is still used despite lower performances as the number of tokens increases.
The code is however merged as is, and #332 is created to investigate the issues.