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

port deepseekv2 and mtp to main branch (#429)

Browse Source 
### What this PR does / why we need it?
This PR ports all the deepseek graph mode code and mtp code from v0.7.3
to the main branch
---------

Signed-off-by: SidaoY <1024863041@qq.com>
Signed-off-by: linfeng-yuan <1102311262@qq.com>
Signed-off-by: Yizhou Liu <liuyizhou5@h-partners.com>
Signed-off-by: mengwei805 <mengwei25@huawei.com>
Signed-off-by: libaokui <libaokui@huawei.com>
Signed-off-by: q00832892 <qiaoyang19@huawei.com>
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
Co-authored-by: SidaoY <1024863041@qq.com>
Co-authored-by: linfeng-yuan <1102311262@qq.com>
Co-authored-by: Yizhou Liu <liuyizhou5@h-partners.com>
Co-authored-by: mengwei805 <mengwei25@huawei.com>
Co-authored-by: libaokui <libaokui@huawei.com>
This commit is contained in:
Pleaplusone
2025-04-19 17:38:18 +08:00
committed by GitHub
parent 086423dc35
commit 1a1f9a6d89
33 changed files with 3361 additions and 315 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm_ascend/worker/worker_v1.py
View File

@@ -38,6 +38,7 @@ from vllm.v1.outputs import ModelRunnerOutput
from vllm.v1.utils import bind_kv_cache
from vllm.v1.worker.worker_base import WorkerBase
from vllm_ascend.distributed.parallel_state import init_ascend_model_parallel
from vllm_ascend.platform import NPUPlatform
from vllm_ascend.utils import try_register_lib, vllm_version_is
from vllm_ascend.worker.model_runner_v1 import NPUModelRunner
@@ -209,6 +210,8 @@ class NPUWorker(WorkerBase):
def _init_worker_distributed_environment(self) -> None:
"""Initialize the distributed environment."""
additional_config = self.vllm_config.additional_config
parallel_config = self.vllm_config.parallel_config
set_custom_all_reduce(
not self.parallel_config.disable_custom_all_reduce)
init_distributed_environment(self.parallel_config.world_size,
@@ -217,6 +220,13 @@ class NPUWorker(WorkerBase):
ensure_model_parallel_initialized(
self.parallel_config.tensor_parallel_size,
self.parallel_config.pipeline_parallel_size)
expert_tensor_parallel_size = 1
if additional_config is not None and "expert_tensor_parallel_size" in additional_config:
expert_tensor_parallel_size = int(
additional_config["expert_tensor_parallel_size"])
init_ascend_model_parallel(parallel_config.tensor_parallel_size,
parallel_config.pipeline_parallel_size,
expert_tensor_parallel_size)
ensure_kv_transfer_initialized(self.vllm_config)
def _init_profiler(self):