EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[misc] Remove useless patch_logits (#4252)

Browse Source 
Torch-npu 2.7.1 has fixed the device check bug. This patch can be
removed now.

- vLLM main:
2918c1b49c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-11-25 21:25:54 +08:00
committed by GitHub
parent 4864909648
commit 98031653df
3 changed files with 1 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
vllm_ascend/patch/worker/__init__.py
View File

@@ -23,7 +23,6 @@ if HAS_TRITON:
# isort: off
import vllm_ascend.patch.platform.patch_sched_yield # noqa
import vllm_ascend.patch.worker.patch_distributed # noqa
import vllm_ascend.patch.worker.patch_logits # noqa
import vllm_ascend.patch.worker.patch_roberta # noqa
import vllm_ascend.patch.worker.patch_weight_loader # noqa
import vllm_ascend.patch.worker.patch_multimodal_merge # noqa

26
vllm_ascend/patch/worker/patch_logits.py
View File

@@ -1,26 +0,0 @@
import torch
import vllm
from vllm._custom_ops import apply_repetition_penalties_torch
def apply_repetition_penalties(logits: torch.Tensor, prompt_mask: torch.Tensor,
output_mask: torch.Tensor,
repetition_penalties: torch.Tensor) -> None:
"""Apply repetition penalties to logits in-place.
Args:
logits: The logits tensor of shape [num_seqs, vocab_size].
prompt_mask: A boolean tensor indicating which tokens appear in the prompt.
output_mask: A boolean tensor indicating which tokens appear in the output.
repetition_penalties: The repetition penalties of shape (num_seqs, ).
"""
apply_repetition_penalties_torch(logits, prompt_mask, output_mask,
repetition_penalties)
# NPU device type tensors have attributes is_cuda=True and is_npu=True, according to its implementation in
# https://github.com/Ascend/pytorch/blob/863b9071cbdf47023c12c246e3efa9c6e2285fc6/torch_npu/npu/_stream_check.py#L74
# This causes that vLLM's apply_repetition_penalties function will run into the branch of "if logits.is_cuda" and
# call the custom op implemented in CUDA, which is not compatible with NPU.
# Reference: https://github.com/vllm-project/vllm/blob/f66673a39d9f364194c249f28098cad8a5584ccb/vllm/_custom_ops.py#L314
vllm._custom_ops.apply_repetition_penalties = apply_repetition_penalties