adds triton flash attention2 kernel #4337
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
stephen-youn
requested review from
RezaYazdaniAminabadi,
jeffra,
mrwyattii,
awan-10,
cmikeh2,
arashb and
tjruwase
as code owners
lekurile
requested changes
Sep 14, 2023
25 tasks
CurryRice233
pushed a commit
to CurryRice233/DeepSpeed
that referenced
this pull request
Sep 28, 2023
* origin/master: Allow multiple inference engines in single script (microsoft#4384) adds triton flash attention2 kernel (microsoft#4337) Fix llama meta tensor loading in AutoTP and kernel injected inference (microsoft#3608) Fix min torch version (microsoft#4375) Fix multinode runner to properly append to PDSH_SSH_ARGS_APPEND (microsoft#4373) add the missing method (microsoft#4363) Openfold fix (microsoft#4368) deepspeed4science japanese blog (microsoft#4369) deepspeed4science chinese blog (microsoft#4366) Enable workflow dispatch on Torch 1.10 CI tests (microsoft#4361) Update conda env to have max pydantic version (microsoft#4362) add deepspeed4science blog link (microsoft#4364) added check to avoid undefined behavior when the input_id length is greater than max_tokens (microsoft#4349) Add the policy to run llama model from the official repo (microsoft#4313) fix deepspeed4science links (microsoft#4358) DeepSpeed4Science (microsoft#4357) Support InternLM (microsoft#4137) Pass base_dir to model files can be loaded for auto-tp/meta-tensor. (microsoft#4348)
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Outdated suggestions cannot be applied.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.
This PR adds flash attention 2 kernel implemented in triton2.1 for inference.
Benchmarking on bert-base shows 4~13% further latency reduction (the longer the sequence length, the larger the performance gain) when compared to the case using regular attention triton kernel.