Skip to content

Fix some speculative decode tests with tl.dot #17371

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 30, 2025
Merged

Fix some speculative decode tests with tl.dot #17371

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8016494
Fix some spec decode tests with tl.dot
huydhn Apr 29, 2025
8b47659
Another
huydhn Apr 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 3 additions & 6 deletions tests/spec_decode/e2e/test_multistep_correctness.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,7 +456,7 @@ def test_spec_decode_e2e_greedy_correctness_real_model_large_bs(
@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,
Expand Down Expand Up @@ -526,11 +526,8 @@ def test_spec_decode_e2e_greedy_correctness_with_preemption(
@pytest.mark.parametrize(
"per_test_common_llm_kwargs",
[
# As of this writing, vLLM only compiles with these 3 block sizes by
# default.
{
"block_size": 8,
},
# https://github.com/triton-lang/triton/issues/2266 tl.dot
# doesn't support embedding < 16
{
"block_size": 16,
},
Expand Down