apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-09-29 05:42:27 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

[Feature] support ep in mixed mode (#3001)

Browse Source 
* [LLM] support ep

* Update worker_process.py

* Update expert_service.py

* Update worker_process.py

* format files
This commit is contained in:
ltd0924
2025-07-30 20:43:39 +08:00
committed by GitHub
parent bd29b2aaca
commit d17886de19
4 changed files with 58 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

56
fastdeploy/engine/engine.py
View File

@@ -243,38 +243,38 @@ class LLMEngine:
self.splitwise_receive_thread.daemon = True
self.splitwise_receive_thread.start()
self.cfg.init_cache_info()
self.cfg.init_cache_info()
role = self.cfg.splitwise_role
host_ip = self.cfg.host_ip
disaggregate = self.cfg.disaggregate_info
if self.cfg.scheduler_config.name == "splitwise":
self.scheduler.start(role, host_ip, disaggregate)
role = self.cfg.splitwise_role
host_ip = self.cfg.host_ip
disaggregate = self.cfg.disaggregate_info
if self.cfg.scheduler_config.name == "splitwise":
self.scheduler.start(role, host_ip, disaggregate)
time.sleep(1)
time.sleep(1)
if self.cfg.parallel_config.enable_expert_parallel and self.cfg.parallel_config.data_parallel_size > 1:
self.dp_processed = []
for i in range(
1,
self.cfg.parallel_config.data_parallel_size // self.cfg.nnode,
):
time.sleep(1)
self.dp_processed.append(
multiprocessing.Process(
target=start_expert_service,
args=(
self.cfg,
i + self.cfg.node_rank * self.cfg.worker_num_per_node,
self.ipc_signal_suffix,
),
)
if self.cfg.parallel_config.enable_expert_parallel and self.cfg.parallel_config.data_parallel_size > 1:
self.dp_processed = []
for i in range(
1,
self.cfg.parallel_config.data_parallel_size // self.cfg.nnode,
):
time.sleep(1)
self.dp_processed.append(
multiprocessing.Process(
target=start_expert_service,
args=(
self.cfg,
i + self.cfg.node_rank * self.cfg.worker_num_per_node,
self.ipc_signal_suffix,
),
)
llm_logger.info(
f"Engine is initialized successfully with {self.cfg.tensor_parallel_size}"
+ f" data parallel id {i}"
)
self.dp_processed[-1].start()
)
llm_logger.info(
f"Engine is initialized successfully with {self.cfg.tensor_parallel_size}"
+ f" data parallel id {i}"
)
self.dp_processed[-1].start()
console_logger.info(f"Worker processes are launched with {time.time() - start_time} seconds.")
return True