wassname/vllm
1
0
Fork 0
mirror of https://github.com/wassname/vllm.git synced 2026-09-10 12:41:29 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Kernel] Replaced blockReduce[...] functions with cub::BlockReduce (#7233)

Browse Source 
Co-authored-by: Michael Goin <michael@neuralmagic.com>
This commit is contained in:
Luka Govedič
2024-08-21 20:18:00 -04:00
committed by GitHub
co-authored by Michael Goin
parent 9984605412
commit 7937009a7e
8 changed files with 237 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/basic_correctness/test_chunked_prefill.py
+1 -1
View File
@@ -83,7 +83,7 @@ def test_models(
for m in E4M3_KV_MODELS])
# Due to low-precision numerical divergence, we only test logprob of 4 tokens
@pytest.mark.parametrize("max_tokens", [4])
@pytest.mark.parametrize("chunked_prefill_token_size", [1, 4, 16])
@pytest.mark.parametrize("chunked_prefill_token_size", [4, 16])
@pytest.mark.parametrize("enforce_eager", [False, True])
# NOTE: Increasing this in this suite will fail CI because we currently cannot
# reset distributed env properly. Use a value > 1 just when you test.