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

[refactor] Remove unnecessary attributes from set_ascend_forward_context (#5204)

Browse Source 
### What this PR does / why we need it?
Remove unnecessary attributes from set_ascend_forward_context
1.prefetch_stream
2.weight_prefetch_method
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: Wang Kunpeng <1289706727@qq.com>
This commit is contained in:
Wang Kunpeng
2025-12-23 08:49:52 +08:00
committed by GitHub
parent 95e8a52156
commit c3a8d13ca7
10 changed files with 55 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_ascend/ops/fused_moe/experts_selector.py
View File

@@ -18,7 +18,8 @@ from typing import Callable, Optional
import torch
import torch_npu
from vllm.forward_context import get_forward_context
from vllm_ascend.utils import get_weight_prefetch_method
def select_experts(hidden_states: torch.Tensor,
@@ -56,7 +57,7 @@ def select_experts(hidden_states: torch.Tensor,
topk_ids: selected expert IDs of shape (num_tokens, top_k).
"""
# prefetch w1_w3_proj.weight preprocess
weight_prefetch_method = get_forward_context().weight_prefetch_method
weight_prefetch_method = get_weight_prefetch_method()
if weight_prefetch_method:
weight_prefetch_method.maybe_prefetch_moe_weight_preprocess(
hidden_states, "gate_up")

5
vllm_ascend/ops/fused_moe/moe_mlp.py
View File

@@ -24,7 +24,8 @@ from vllm.triton_utils import HAS_TRITON
from vllm_ascend.ascend_forward_context import MoECommType
from vllm_ascend.utils import (AscendDeviceType, dispose_tensor,
enable_custom_op, get_ascend_device_type)
enable_custom_op, get_ascend_device_type,
get_weight_prefetch_method)
def _custom_gmm_swiglu_enabled(fusion, dynamic_eplb):
@@ -100,7 +101,7 @@ def quant_apply_mlp(hidden_states: torch.Tensor,
bias1, bias2 = None, None
_output_dtype = w2_scale[0].dtype
weight_prefetch_method = get_forward_context().weight_prefetch_method
weight_prefetch_method = get_weight_prefetch_method()
if weight_prefetch_method:
weight_prefetch_method.maybe_prefetch_moe_weight_postprocess(
hidden_states)