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

[Refactor] move the metadata from attention_v1 to util(ready for extract common_cp) & realize Ascendmetadata inherit from the parent class. (#5203)

Browse Source 
RFC: https://github.com/vllm-project/vllm-ascend/issues/4629

1. Remove the pcp-related code from attention_v1.
2. Establish the inheritance relationship of CommonAttentionMetadata.

TODO
1. extract common_cp
2. move cp metadata to common_cp.
3. remove commonAttentionMetadata for aclgraph.

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

---------

Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>
This commit is contained in:
weijinqian0
2025-12-23 00:10:52 +08:00
committed by GitHub
parent 3d9954eff0
commit 95e8a52156
9 changed files with 83 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm_ascend/spec_decode/mtp_proposer.py
View File

@@ -268,7 +268,7 @@ class MtpProposer(Proposer):
spec_attn_mask=self.runner.spec_attn_mask,
attn_state=self.runner.attn_state,
decode_token_per_req=self.runner.decode_token_per_req,
)
max_seq_len=0)
if self.pcp_size * self.dcp_size > 1:
# update long_seq related params and flatten block_table
common_attn_metadata.prefill_context_parallel_metadata = \
@@ -599,7 +599,7 @@ class MtpProposer(Proposer):
spec_attn_mask=self.runner.spec_attn_mask,
attn_state=self.runner.attn_state,
decode_token_per_req=self.runner.decode_token_per_req,
)
max_seq_len=0)
return spec_common_attn_metadata, token_indices
def _propose(
@@ -1221,7 +1221,8 @@ class MtpProposer(Proposer):
decode_token_per_req=self.runner.decode_token_per_req,
num_computed_tokens_cpu=common_attn_metadata.
num_computed_tokens_cpu,
seq_lens=common_attn_metadata.seq_lens)
seq_lens=common_attn_metadata.seq_lens,
max_seq_len=0)
query_start_loc = common_attn_metadata.query_start_loc[
1:1 + num_rejected_tokens_gpu.shape[0]]