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

[feature]dcp&pcp support mlapo (#5672)

Browse Source 
### What this PR does / why we need it?
mlapo in deepseek is a huge performance improvement in decode, this pr
support pcp & dcp with mlapo

### Does this PR introduce _any_ user-facing change?
NO

### How was this patch tested?

- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef

---------

Signed-off-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
This commit is contained in:
zhenwenqi2024
2026-01-08 23:49:23 +08:00
committed by GitHub
parent 6315a31399
commit 97f6be8108
7 changed files with 35 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm_ascend/worker/model_runner_v1.py
View File

@@ -912,12 +912,15 @@ class NPUModelRunner(GPUModelRunner):
self.input_batch)
blk_table.slot_mapping.gpu[maybe_pcp_full_tokens:].fill_(-1)
if self.pcp_size > 1:
slot_mapping = self.pcp_manager.get_padded_slot_mapping(
slot_mapping_pcp = self.pcp_manager.get_padded_slot_mapping(
total_num_scheduled_tokens,
slot_mapping,
)
blk_table.slot_mapping.gpu[:self.pcp_manager.
num_actual_tokens_pcp_padded] = slot_mapping
num_actual_tokens_pcp_padded] = slot_mapping_pcp
slot_mapping = blk_table.slot_mapping.gpu[:self.
pcp_manager.
num_actual_tokens_pcp_padded]
# NOTE: This is a temporary hack, now in GPUModelRunner, this prepare_inputs
# has been split to multiple parts, and there are 3 parts that is related to this