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

chore: upgrade sgl-kernel v0.1.8.post2 (#7186)

Browse Source 
Co-authored-by: zhyncs <me@zhyncs.com>
This commit is contained in:
JieXin Liang
2025-06-15 09:26:18 +08:00
committed by GitHub
parent 55561e2553
commit ed89837cf4
4 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/pyproject.toml
View File

@@ -49,7 +49,7 @@ runtime_common = [
srt = [ srt = [
"sglang[runtime_common]", "sglang[runtime_common]",
"sgl-kernel==0.1.8.post1", "sgl-kernel==0.1.8.post2",
"flashinfer_python==0.2.6.post1", "flashinfer_python==0.2.6.post1",
"torch==2.7.1", "torch==2.7.1",
"torchaudio==2.7.1", "torchaudio==2.7.1",

2
python/sglang/srt/entrypoints/engine.py
View File

@@ -605,7 +605,7 @@ def _set_envs_and_config(server_args: ServerArgs):
if _is_cuda: if _is_cuda:
assert_pkg_version( assert_pkg_version(
"sgl-kernel", "sgl-kernel",
"0.1.8.post1", "0.1.8.post2",
"Please reinstall the latest version with `pip install sgl-kernel --force-reinstall`", "Please reinstall the latest version with `pip install sgl-kernel --force-reinstall`",
) )

1
python/sglang/srt/layers/attention/cutlass_mla_backend.py
View File

@@ -280,6 +280,7 @@ class CutlassMLABackend(FlashInferMLAAttnBackend):
seq_lens=forward_batch.seq_lens.to(torch.int32), seq_lens=forward_batch.seq_lens.to(torch.int32),
page_table=self.forward_metadata.block_kv_indices, page_table=self.forward_metadata.block_kv_indices,
workspace=self.forward_metadata.workspace, workspace=self.forward_metadata.workspace,
sm_scale=layer.scaling,
num_kv_splits=1, num_kv_splits=1,
) )

6
python/sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py
View File

@@ -52,7 +52,11 @@ def grouped_gemm_nt_f8f8bf16_masked(
expected_m, n, k, num_groups, kernel_type expected_m, n, k, num_groups, kernel_type
): ):
_grouped_gemm_nt_f8f8bf16_masked_raw( _grouped_gemm_nt_f8f8bf16_masked_raw(
lhs, rhs, out, masked_m, expected_m, lhs,
rhs,
out,
masked_m,
expected_m,
**({"recipe": recipe} if DEEPGEMM_V202506 else {}) **({"recipe": recipe} if DEEPGEMM_V202506 else {})
) )