[Spec Decoding] Add DFlash e2e tests and Buildkite CI#1870
Open
aaronzhfeng wants to merge 1 commit intovllm-project:mainfrom
Open
[Spec Decoding] Add DFlash e2e tests and Buildkite CI#1870aaronzhfeng wants to merge 1 commit intovllm-project:mainfrom
aaronzhfeng wants to merge 1 commit intovllm-project:mainfrom
Conversation
Signed-off-by: aaronzhfeng <fzx333578@gmail.com>
aaronzhfeng
requested review from
QiliangCui,
jcyang43 and
vipannalla
as code owners
3 tasks
Collaborator
|
Thanks so much for this contribution and breaking up the PRs like this, once the other 2 are in good shape, could you please review this guidance for landing the feature in our support matrices: |
Collaborator
|
Consider adding the tests to pipeline_jax.yml so that it can run before merging every PR. |
aaronzhfeng
added a commit
to aaronzhfeng/tpu-inference
that referenced
this pull request
Apr 10, 2026
…ject#1869 + vllm-project#1870) Integration from dev workspace for full e2e validation: - tpu_runner.py: DFlash + DFlash torchax dispatch - speculative_decoding_manager.py: DFlash propose flow - model_loader.py: DFlash model registration - compilation_manager.py: DFlash precompilation - test_speculative_decoding.py: e2e test Signed-off-by: Aaron Feng <aaronzhfeng@gmail.com> Signed-off-by: aaronzhfeng <fzx333578@gmail.com>
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
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Add e2e tests and Buildkite CI for DFlash block-diffusion speculative decoding. The DFlash model/proposer were added in #1868, and pipeline integration in #1869. This PR adds the test coverage and CI.
Verified on both TPU v4 and v5p across 9 datasets (math, code, chat) with Qwen3-4B target + z-lab/Qwen3-4B-DFlash-b16 draft, achieving 3x average speedup.
Files:
tests/e2e/test_speculative_decoding.py-- addtest_dflash_correctness(Qwen3-4B + DFlash draft, output correctness) andtest_dflash_performance(1.5x speedup threshold).buildkite/features/Speculative_Decoding-_DFlash.yml-- Buildkite CI pipeline for DFlash correctness and performance, modeled after Eagle3'sSpeculative_Decoding-_Eagle3.ymlTests
Checklist