biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI Fix] Try fixing eagle e2e test OOM by reducing block allocation (#20213)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-06-29 11:31:37 +09:00
committed by GitHub
parent 4d36693687
commit 7b1895e6ce
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/spec_decode/e2e/test_eagle_correctness.py
View File

@@ -370,6 +370,10 @@ def test_llama2_eagle_e2e_greedy_correctness(vllm_runner, common_llm_kwargs,
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
# 2 for small prompt, 256//16 for generated.
"num_gpu_blocks_override": 2 + 256 // 16,
"max_model_len": (2 + 256 // 16) * 16,
# Skip cuda graph recording for fast test.
"enforce_eager": True,
@@ -420,6 +424,10 @@ def test_llama3_eagle_e2e_greedy_correctness(vllm_runner, common_llm_kwargs,
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
# 2 for small prompt, 256//16 for generated.
"num_gpu_blocks_override": 2 + 256 // 16,
"max_model_len": (2 + 256 // 16) * 16,
# Skip cuda graph recording for fast test.
"enforce_eager": True,