biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Feature] Prefill Context Parallel (PCP) basic support (#28718)

Browse Source 
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Signed-off-by: FENP <yuanyongjie.yyj@antgroup.com>
Signed-off-by: LookAround <lixushi@huawei.com>
Signed-off-by: Jingchun Gao <gaojingchun1@huawei.com>
Signed-off-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Co-authored-by: FENP <yuanyongjie.yyj@antgroup.com>
Co-authored-by: LookAround <lixushi@huawei.com>
Co-authored-by: Jingchun Gao <gaojingchun1@huawei.com>
Co-authored-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Co-authored-by: Jingchun Gao <63247409+gjc0824@users.noreply.github.com>
This commit is contained in:
Qiu
2025-11-20 04:52:44 +08:00
committed by GitHub
parent 02f5903b84
commit 2fd893b4ce
27 changed files with 399 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/models/gpt_oss.py
View File

@@ -13,6 +13,7 @@ from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import (
get_dp_group,
get_ep_group,
get_pcp_group,
get_pp_group,
get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size,
@@ -322,10 +323,12 @@ class GptOssModel(nn.Module):
# In MoE, we need to flatten the tensor parallel size across the data
# parallel size when EP is disabled.
tp_size, tp_rank = FusedMoEParallelConfig.flatten_tp_across_dp(
tp_size, tp_rank = FusedMoEParallelConfig.flatten_tp_across_dp_and_pcp(
tp_size=get_tensor_model_parallel_world_size(),
dp_size=get_dp_group().world_size,
dp_rank=get_dp_group().rank_in_group,
pcp_size=get_pcp_group().world_size,
pcp_rank=get_pcp_group().rank_in_group,
)
intermediate_size = self.config.intermediate_size
@@ -507,10 +510,12 @@ class GptOssModel(nn.Module):
# In MoE, we need to flatten the tensor parallel size across the data
# parallel size when EP is disabled.
tp_size, tp_rank = FusedMoEParallelConfig.flatten_tp_across_dp(
tp_size, tp_rank = FusedMoEParallelConfig.flatten_tp_across_dp_and_pcp(
tp_size=get_tensor_model_parallel_world_size(),
dp_size=get_dp_group().world_size,
dp_rank=get_dp_group().rank_in_group,
pcp_size=get_pcp_group().world_size,
pcp_rank=get_pcp_group().rank_in_group,
)
intermediate_size = self.config.intermediate_size