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

[Bug] Fix duplicate 'torch.' prefix in qwen-vl (#1986)

Browse Source 
Signed-off-by: wuzhongjian <wuzhongjian_yewu@cmss.chinamobile.com>

### What this PR does / why we need it?
Fix duplicate 'torch.' prefix in qwen2-vl, qwen2.5-vl

- vLLM version: v0.9.2
- vLLM main:
dde295a934
This commit is contained in:
JohnJan
2025-07-24 20:16:00 +08:00
committed by GitHub
parent 84fc7402c3
commit cfdd45ed00
3 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm_ascend/models/qwen2_5_vl.py
View File

@@ -107,7 +107,7 @@ class AscendQwen2_5_VisionAttention(Qwen2_5_VisionAttention):
for x in (q, k, v) for x in (q, k, v)
] ]
context_layer = torch.torch.empty_like(q) context_layer = torch.empty_like(q)
# operator requires pta version >= 2.5.1 # operator requires pta version >= 2.5.1
torch_npu._npu_flash_attention_unpad( torch_npu._npu_flash_attention_unpad(

2
vllm_ascend/models/qwen2_5_vl_without_padding.py
View File

@@ -87,7 +87,7 @@ class AscendQwen2_5_VisionAttention_Without_Padding(Qwen2_5_VisionAttention):
for x in (q, k, v) for x in (q, k, v)
] ]
context_layer = torch.torch.empty_like(q) context_layer = torch.empty_like(q)
# operator requires pta version >= 2.5.1.dev20250226 # operator requires pta version >= 2.5.1.dev20250226
torch_npu._npu_flash_attention_unpad( torch_npu._npu_flash_attention_unpad(

2
vllm_ascend/models/qwen2_vl.py
View File

@@ -95,7 +95,7 @@ class AscendQwen2VisionAttention(Qwen2VisionAttention):
for x in (q, k, v) for x in (q, k, v)
] ]
context_layer = torch.torch.empty_like(q) context_layer = torch.empty_like(q)
# operator requires pta version >= 2.5.1 # operator requires pta version >= 2.5.1
torch_npu._npu_flash_attention_unpad( torch_npu._npu_flash_attention_unpad(