minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[NPU] fix model llava-onevision-qwen2-7b-ov torch compiles error in npu case (#26573)

Browse Source
This commit is contained in:
ZeyuanChen2000
2026-05-30 17:10:58 +08:00
committed by GitHub
parent 714bcd84e2
commit 90eb894564
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/sampling/penaltylib/repetition_penalty.py
+4 -2
View File
@@ -1,10 +1,12 @@
import torch import torch
from sglang.srt.sampling.penaltylib.orchestrator import _BatchedPenalizer from sglang.srt.sampling.penaltylib.orchestrator import _BatchedPenalizer
from sglang.srt.utils import get_compiler_backend from sglang.srt.utils import get_compiler_backend, is_npu
_is_npu = is_npu()
@torch.compile(dynamic=True, backend=get_compiler_backend()) @torch.compile(dynamic=True, backend=get_compiler_backend(), disable=_is_npu)
def apply_scaling_penalties(logits, scaling_penalties): def apply_scaling_penalties(logits, scaling_penalties):
logits[:] = torch.where( logits[:] = torch.where(
logits < 0, logits < 0,