xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:45:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix more broken speculative decode tests (#17450)

Browse Source 
Signed-off-by: Huy Do <huydhn@gmail.com>
This commit is contained in:
Huy Do 2025-05-01 06:05:58 -07:00 committed by GitHub
parent 2007d4d54f
commit b74d888c63
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 9 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/spec_decode/e2e/test_medusa_correctness.py
View File

@ -205,7 +205,7 @@ def test_medusa_e2e_greedy_correctness_cuda_graph(
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"block_size": 8,
"block_size": 16,
# 2 for small prompt, 256//8 for generated.
"num_gpu_blocks_override": 2 + 256 // 8,
"max_model_len": (2 + 256 // 8) * 8,

4
tests/spec_decode/e2e/test_mlp_correctness.py
View File

@ -267,7 +267,7 @@ def test_mlp_e2e_seeded_correctness(vllm_runner, common_llm_kwargs,
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"block_size": 8,
"block_size": 16,
# 2 for small prompt, 256//8 for generated.
"num_gpu_blocks_override": 2 + 256 // 8,
"max_model_len": (2 + 256 // 8) * 8,
@ -321,7 +321,7 @@ def test_mlp_e2e_greedy_correctness_with_preemption(
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"block_size": 8,
"block_size": 16,
# 2 for small prompt, 256//8 for generated.
"num_gpu_blocks_override": 2 + 256 // 8,
"max_model_len": (2 + 256 // 8) * 8,

2
tests/spec_decode/e2e/test_ngram_correctness.py
View File

@ -152,7 +152,7 @@ def test_ngram_e2e_greedy_logprobs(vllm_runner, common_llm_kwargs,
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"block_size": 8,
"block_size": 16,
# 2 for small prompt, 256//8 for generated.
"num_gpu_blocks_override": 2 + 256 // 8,
"max_model_len": (2 + 256 // 8) * 8,

5
vllm/spec_decode/multi_step_worker.py
View File

@ -51,9 +51,14 @@ class MultiStepWorker(ProposerWorkerBase, DelegateWorkerBase):
def set_include_gpu_probs_tensor(self) -> None:
# Need include_gpu_probs_tensor for MultiStepWorker
self.model_runner.sampler.include_gpu_probs_tensor = True
if hasattr(self.model_runner.model, "sampler"):
(self.model_runner.model.sampler.include_gpu_probs_tensor) = True
def set_should_modify_greedy_probs_inplace(self) -> None:
self.model_runner.sampler.should_modify_greedy_probs_inplace = True
if hasattr(self.model_runner.model, "sampler"):
(self.model_runner.model.sampler.should_modify_greedy_probs_inplace
) = True
@torch.inference_mode()
def sampler_output(