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

23
vllm_ascend/distributed/mooncake_connector.py
View File

@@ -27,9 +27,10 @@ from vllm.distributed.kv_transfer.kv_connector.v1.base import (
KVConnectorBase_V1, KVConnectorMetadata, KVConnectorRole)
from vllm.distributed.parallel_state import (
get_decode_context_model_parallel_rank,
get_decode_context_model_parallel_world_size,
get_decode_context_model_parallel_world_size, get_pcp_group,
get_tensor_model_parallel_rank, get_tp_group)
from vllm.logger import logger
from vllm.utils.network_utils import get_ip, make_zmq_path, make_zmq_socket
from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.kv_cache_interface import KVCacheConfig
from vllm.v1.request import RequestStatus
@@ -38,16 +39,6 @@ import vllm_ascend.envs as envs_ascend
from vllm_ascend.ascend_config import get_ascend_config, init_ascend_config
from vllm_ascend.distributed.mooncake_transfer_engine import global_te
from vllm_ascend.distributed.utils import get_transfer_timeout_value
from vllm_ascend.utils import prefill_context_parallel_enable
# isort: off
if prefill_context_parallel_enable():
from vllm.distributed import (get_prefill_context_model_parallel_rank,
get_prefill_context_model_parallel_world_size
)
# isort: on
from vllm.utils.network_utils import get_ip, make_zmq_path, make_zmq_socket
if TYPE_CHECKING:
from vllm.attention.backends.abstract import AttentionMetadata
@@ -730,8 +721,7 @@ class MooncakeConnectorScheduler:
logger.info("Initializing Mooncake Scheduler %s", engine_id)
self.side_channel_host = get_ip()
self.pcp_size = vllm_config.parallel_config.prefill_context_parallel_size \
if prefill_context_parallel_enable() else 1
self.pcp_size = vllm_config.parallel_config.prefill_context_parallel_size
self.dcp_size = vllm_config.parallel_config.decode_context_parallel_size
self.max_device_id = vllm_config.parallel_config.tensor_parallel_size * \
vllm_config.parallel_config.data_parallel_size * \
@@ -898,10 +888,9 @@ class MooncakeConnectorWorker:
self.dp_size = vllm_config.parallel_config.data_parallel_size_local
self.kv_caches: dict[str, torch.Tensor] = {}
self.side_channel_host = get_ip()
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