[EP] fix several bugs in data parallel (#4657)

* Simplify profiling block setup in expert_service.py

Refactor profiling block initialization to avoid duplication.

* Update common_engine.py
This commit is contained in:
ltd0924
2025-10-30 09:50:49 +08:00
committed by GitHub
parent dab04ab413
commit 50be19a88a
2 changed files with 17 additions and 17 deletions

View File

@@ -303,7 +303,8 @@ class EngineService:
client_id=0,
local_data_parallel_size=self.cfg.parallel_config.data_parallel_size,
local_data_parallel_id=min(
self.cfg.worker_num_per_node * self.cfg.node_rank + self.cfg.parallel_config.local_data_parallel_id,
self.cfg.worker_num_per_node // self.cfg.parallel_config.tensor_parallel_size * self.cfg.node_rank
+ self.cfg.parallel_config.local_data_parallel_id,
self.cfg.parallel_config.data_parallel_size - 1,
),
)