xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-15 16:26:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix some speculative decode tests with tl.dot (#17371)

Browse Source 
Signed-off-by: Huy Do <huydhn@gmail.com>
This commit is contained in:
Huy Do 2025-04-29 19:41:02 -07:00 committed by GitHub
parent d1f569b1b9
commit 88fcf00dda
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
tests/spec_decode/e2e/test_multistep_correctness.py
View File

@ -456,7 +456,7 @@ def test_spec_decode_e2e_greedy_correctness_real_model_large_bs(
@pytest.mark.parametrize( @pytest.mark.parametrize(
"common_llm_kwargs", "common_llm_kwargs",
[{ [{
"block_size": 8, "block_size": 16,
# 2 for small prompt, 256//8 for generated. # 2 for small prompt, 256//8 for generated.
"num_gpu_blocks_override": 2 + 256 // 8, "num_gpu_blocks_override": 2 + 256 // 8,
"max_model_len": (2 + 256 // 8) * 8, "max_model_len": (2 + 256 // 8) * 8,
@ -526,11 +526,8 @@ def test_spec_decode_e2e_greedy_correctness_with_preemption(
@pytest.mark.parametrize( @pytest.mark.parametrize(
"per_test_common_llm_kwargs", "per_test_common_llm_kwargs",
[ [
# As of this writing, vLLM only compiles with these 3 block sizes by # https://github.com/triton-lang/triton/issues/2266 tl.dot
# default. # doesn't support embedding < 16
{
"block_size": 8,
},
{ {
"block_size": 16, "block_size": 16,
}, },