[V1][Spec Decode] Optimize Rejection Sampler with Triton Kernels #14930
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
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
youkaichao
reviewed
Mar 19, 2025
1 task
This was referenced Mar 21, 2025
Closed
1 task
lulmer
pushed a commit
to lulmer/vllm
that referenced
this pull request
Apr 7, 2025
…m-project#14930) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Signed-off-by: Louis Ulmer <ulmerlouis@gmail.com>
shreyankg
pushed a commit
to shreyankg/vllm
that referenced
this pull request
May 3, 2025
…m-project#14930) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
RichardoMrMu
pushed a commit
to RichardoMrMu/vllm
that referenced
this pull request
May 12, 2025
…m-project#14930) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>
mmyxym
reviewed
Aug 5, 2025
| GREEDY_TEMPERATURE: tl.constexpr = -1 | ||
| # Maximum number of speculative draft tokens allowed per request in a single | ||
| # step. This value is chosen to be large enough to handle typical use cases. | ||
| MAX_SPEC_LEN = 32 |
There was a problem hiding this comment.
Hi @WoosukKwon , is there any limitation MAX_SPEC_LEN should be 32? Can it be larger? Thanks.
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Outdated suggestions cannot be applied.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 optimizes the rejection sampler in #13933 with custom Triton kernels.
By using the Triton kernels, the PR brings the following benefits:
[num_tokens, vocab_size]for the logits tensors, instead of[batch_size, max_spec_len, vocab_size]. This reduces the GPU memory usage a lot.cat,gather, etc.)Performance benchmark: Llama 3.1 8B, ShareGPT, 1xH100, temperature 0.1
SD config:
--speculative-model "[ngram]" --ngram_prompt_lookup_min 5 --ngram-prompt-lookup-max 5 --num_speculative_tokens 325% throughput increase compared to main w/o SD, and 18% increase compared to main w/ SD.
Accuracy benchmark: GSM8K, Llama 3.1 8B Instruct, 5 shots