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

[long_seq] remove long_seq env (#4660)

Browse Source 
### What this PR does / why we need it?
remove env VLLM_ASCEND_ENABLE_CONTEXT_PARALLEL 

- vLLM version: v0.12.0

---------

Signed-off-by: LookAround <lixushi@huawei.com>
Signed-off-by: ZhangMingWei716 <2894054457@qq.com>
Co-authored-by: ZhangMingWei716 <2894054457@qq.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
LookAround0301
2025-12-05 10:31:49 +08:00
committed by GitHub
parent ea54388e19
commit b32ef53b3b
16 changed files with 230 additions and 176 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm_ascend/distributed/kvpool/pool_worker.py
View File

@@ -6,7 +6,7 @@ import torch
from vllm.config import VllmConfig
from vllm.distributed import (get_decode_context_model_parallel_rank,
get_decode_context_model_parallel_world_size,
get_tensor_model_parallel_rank,
get_pcp_group, get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size)
from vllm.logger import logger
from vllm.v1.core.kv_cache_utils import BlockHash
@@ -22,14 +22,6 @@ from vllm_ascend.distributed.kvpool.config_data import (
from vllm_ascend.distributed.kvpool.kv_transfer import (
KVCacheStoreLayerRecvingThread, KVCacheStoreLayerSendingThread,
KVCacheStoreRecvingThread, KVCacheStoreSendingThread, KVTransferThread)
from vllm_ascend.utils import prefill_context_parallel_enable
if prefill_context_parallel_enable():
# isort: off
from vllm.distributed import (get_prefill_context_model_parallel_rank,
get_prefill_context_model_parallel_world_size
)
# isort: on
backend_map: Dict[str, Type[Backend]] = {
"mooncake": MooncakeBackend,
@@ -57,10 +49,9 @@ class KVPoolWorker:
self.tp_rank = get_tensor_model_parallel_rank()
self.tp_size = get_tensor_model_parallel_world_size()
self.pcp_size = get_prefill_context_model_parallel_world_size(
) if prefill_context_parallel_enable() else 1
self.pcp_rank = get_prefill_context_model_parallel_rank(
) if self.pcp_size > 1 else 0
self.pcp_size = get_pcp_group().world_size
self.pcp_rank = get_pcp_group(
).rank_in_group if self.pcp_size > 1 else 0
self.dcp_size = get_decode_context_model_parallel_world_size()
self.dcp_rank = get_decode_context_model_parallel_rank(
) if self.dcp_size > 1 else 0