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

Add mrope op fusion (#3509)

Browse Source 
### What this PR does / why we need it?
Add mrope fusion op for qwen2.5-vl. This mrope operator dosen't support
Qwen3-VL currently. Thus could only take affect in qwen2.5-vl

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: shaopeng666 <shaopeng666@noreply.gitcode.com>
Co-authored-by: shaopeng666 <shaopeng666@noreply.gitcode.com>
This commit is contained in:
shaopeng-666
2025-10-18 18:08:24 +08:00
committed by GitHub
parent 0777e2f899
commit 646c1db5d7
3 changed files with 123 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_ascend/utils.py
View File

@@ -517,8 +517,8 @@ def register_ascend_customop(vllm_config: Optional[VllmConfig] = None):
AscendReplicatedLinear,
AscendRowParallelLinear)
from vllm_ascend.ops.rotary_embedding import (
AscendDeepseekScalingRotaryEmbedding, AscendRotaryEmbedding,
AscendYaRNRotaryEmbedding)
AscendDeepseekScalingRotaryEmbedding, AscendMRotaryEmbedding,
AscendRotaryEmbedding, AscendYaRNRotaryEmbedding)
from vllm_ascend.ops.vocab_parallel_embedding import (
AscendLogitsProcessor, AscendParallelLMHead,
AscendVocabParallelEmbedding)
@@ -528,6 +528,7 @@ def register_ascend_customop(vllm_config: Optional[VllmConfig] = None):
"QuickGELU": AscendQuickGELU,
"SiluAndMul": AscendSiluAndMul,
"RotaryEmbedding": AscendRotaryEmbedding,
"MRotaryEmbedding": AscendMRotaryEmbedding,
"ColumnParallelLinear": AscendColumnParallelLinear,
"RowParallelLinear": AscendRowParallelLinear,
"YaRNScalingRotaryEmbedding": AscendYaRNRotaryEmbedding,