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

Update torch-npu version to 2.7.1 (#3896)

Browse Source 
### What this PR does / why we need it?
Upgrade torch-npu to the official release version 2.7.1


- vLLM version: v0.11.0
- vLLM main:
83f478bb19

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-10-31 17:16:31 +08:00
committed by GitHub
parent 5f6d1b3323
commit fcc9a0eaeb
15 changed files with 83 additions and 168 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
vllm_ascend/compilation/acl_graph.py
View File

@@ -19,8 +19,6 @@ from vllm.forward_context import BatchDescriptor, get_forward_context
from vllm.logger import logger
from vllm.platforms import current_platform
from vllm_ascend.utils import version_check
from ..utils import weak_ref_tensors
@@ -214,32 +212,20 @@ def update_attn_params(update_stream, forward_context, runtime_shape):
output,
) = param
seq_lens = forward_context.attn_metadata[key].seq_lens
torch_npu_check = version_check()
with torch.npu.stream(update_stream):
torch.npu.graph_task_update_begin(update_stream, handle)
if torch_npu_check:
torch_npu._npu_paged_attention(
query=query,
key_cache=key_cache,
value_cache=value_cache,
num_kv_heads=num_kv_heads,
num_heads=num_heads,
scale_value=scale,
block_table=block_table,
context_lens=seq_lens,
out=output,
workspace=graph_params.workspaces.get(runtime_shape))
else:
torch_npu._npu_paged_attention(query=query,
key_cache=key_cache,
value_cache=value_cache,
num_kv_heads=num_kv_heads,
num_heads=num_heads,
scale_value=scale,
block_table=block_table,
context_lens=seq_lens,
out=output)
torch_npu._npu_paged_attention(
query=query,
key_cache=key_cache,
value_cache=value_cache,
num_kv_heads=num_kv_heads,
num_heads=num_heads,
scale_value=scale,
block_table=block_table,
context_lens=seq_lens,
out=output,
workspace=graph_params.workspaces.get(runtime_shape))
torch.npu.graph_task_update_end(update_stream)
event.record(update_stream)