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

Drop vLLM 0.13.0 support (#6069)

Browse Source 
### What this PR does / why we need it?
Drop vLLM 0.13.0 support, upgrade to 0.14.0

- vLLM version: v0.13.0
- vLLM main:
d68209402d

---------

Signed-off-by: hfadzxy <starmoon_zhang@163.com>
This commit is contained in:
zhangxinyuehfad
2026-01-23 09:45:08 +08:00
committed by GitHub
parent 27a513b672
commit 819a4459ce
39 changed files with 86 additions and 272 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm_ascend/ops/fused_moe/fused_moe.py
View File

@@ -50,7 +50,7 @@ from vllm_ascend.quantization.w8a8_dynamic import \
from vllm_ascend.utils import (AscendDeviceType, enable_sp,
get_ascend_device_type, maybe_trans_nz,
npu_stream_switch, shared_expert_dp_enabled,
shared_experts_calculation_stream, vllm_version_is)
shared_experts_calculation_stream)
@dataclass
class FusedMoEResult:
@@ -451,12 +451,7 @@ class AscendSharedFusedMoE(SharedFusedMoE, AscendFusedMoE):
# Qwen3-Next specific gating mechanism
if hasattr(self._shared_experts, "expert_gate") and \
self._shared_experts.expert_gate is not None:
if vllm_version_is('0.13.0'):
# TODO(jianzs): remove this branch after vLLM new version is
# released
gate_out = self._shared_experts.expert_gate(hidden_states) # type: ignore
else:
gate_out, _ = self._shared_experts.expert_gate(hidden_states) # type: ignore
gate_out, _ = self._shared_experts.expert_gate(hidden_states) # type: ignore
shared_out = F.sigmoid(gate_out) * shared_out
return shared_out