EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

fix file name spelling mistake and useless variable in minmax-text-01-lightning_attention (#2971)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-01-19 10:56:13 +08:00
committed by GitHub
parent 3d93f84a00
commit 83452dbb4a
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
benchmark/kernels/minmax-text-01-lighting_attention/benchmark_lighting_attention_decode.py → benchmark/kernels/minmax-text-01-lightning_attention/benchmark_lightning_attention_decode.py
View File

@@ -26,7 +26,6 @@ def _decode_kernel(
d_original: tl.constexpr, d_original: tl.constexpr,
e: tl.constexpr, e: tl.constexpr,
e_original: tl.constexpr, e_original: tl.constexpr,
BLOCK_SIZE: tl.constexpr = 32,
): ):
off_bh = tl.program_id(0) off_bh = tl.program_id(0)
off_h = off_bh % h off_h = off_bh % h

2
benchmark/kernels/minmax-text-01-lighting_attention/benchmark_lighting_attention_prefill.py → benchmark/kernels/minmax-text-01-lightning_attention/benchmark_lightning_attention_prefill.py
View File

@@ -493,6 +493,8 @@ def test_lightning_attention_implementations(model_params):
msg="Lightning attention implementations produce different results", msg="Lightning attention implementations produce different results",
) )
print("✅ Two implementations match")
def get_benchmark(): def get_benchmark():
batch_size_range = [2**i for i in range(0, 7)] # max 64 batch_size_range = [2**i for i in range(0, 7)] # max 64