karylab_agents/vllm
1
0
Fork 0
forked from Karylab-cklius/vllm
Code Pull Requests 2 Actions 1 Packages Activity

[ROCm] [DSv4] [Perf] Support DeepSeek v4 MTP (#43385)

Browse Source 
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
This commit is contained in:
TJian
2026-05-24 18:43:08 +08:00
committed by GitHub
parent 5940590855
commit 1806d1adfc
6 changed files with 2340 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/model_executor/layers/fused_moe/experts/gpt_oss_triton_kernels_moe.py
+22 -11
View File
@@ -81,17 +81,28 @@ def _patch_make_bitmatrix_metadata() -> None:
import triton.language as tl
try:
from vllm.third_party.triton_kernels.tensor_details import (
bitmatrix as _bm,
)
from vllm.third_party.triton_kernels.tensor_details.bitmatrix import (
BitmatrixMetadata,
_keyed_add,
cdiv,
)
from vllm.third_party.triton_kernels.tensor_details.bitmatrix_details.sum_bitmatrix_rows import ( # noqa: E501
sum_bitmatrix_rows,
)
if current_platform.is_rocm():
from triton_kernels.tensor_details import bitmatrix as _bm
from triton_kernels.tensor_details.bitmatrix import (
BitmatrixMetadata,
_keyed_add,
cdiv,
)
from triton_kernels.tensor_details.bitmatrix_details.sum_bitmatrix_rows import ( # noqa: E501
sum_bitmatrix_rows,
)
else:
from vllm.third_party.triton_kernels.tensor_details import (
bitmatrix as _bm,
)
from vllm.third_party.triton_kernels.tensor_details.bitmatrix import (
BitmatrixMetadata,
_keyed_add,
cdiv,
)
from vllm.third_party.triton_kernels.tensor_details.bitmatrix_details.sum_bitmatrix_rows import ( # noqa: E501
sum_bitmatrix_rows,
)
except ImportError:
return