apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[PD Disaggregation] support DP via v1 router and decouple DP and EP (#5197)

Browse Source 
* [fix] support DP via v1 router and decouple DP and EP

* [fix] fix scripts

* [fix] reset model path

* [fix] dp use get_output_ep, fix router port type, update scripts

* [merge] merge with latest code

* [chore] remove some debug log

* [fix] fix code style check

* [fix] fix test_multi_api_server for log_dir name

* [chore] reduce logs

* Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This commit is contained in:
Yonghua Li
2025-12-04 15:38:43 +08:00
committed by GitHub
parent 5cd17fd662
commit f4119d51b4
15 changed files with 394 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
fastdeploy/worker/worker_process.py
View File

@@ -173,7 +173,11 @@ class PaddleDisWorkerProc:
model_weights_status:
"""
self.max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
if self.parallel_config.data_parallel_size > 1 and not envs.FD_ENABLE_MULTI_API_SERVER:
if (
self.parallel_config.enable_expert_parallel
and self.parallel_config.data_parallel_size > 1
and not envs.FD_ENABLE_MULTI_API_SERVER
):
launched_expert_service_signal_data = np.zeros(
shape=[self.parallel_config.data_parallel_size // self.fd_config.nnode], dtype=np.int32
)
@@ -905,6 +909,12 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
parallel_config.tensor_parallel_rank = local_rank % parallel_config.tensor_parallel_size
parallel_config.data_parallel_rank = local_rank // parallel_config.tensor_parallel_size
# config for DP
if parallel_config.data_parallel_size > 1:
max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
parallel_config.local_data_parallel_id = parallel_config.data_parallel_rank % (
max_chips_per_node // parallel_config.tensor_parallel_size
)
# config for EP
if parallel_config.expert_parallel_size > 1:
expert_parallel_rank = int(local_rank % parallel_config.expert_parallel_size)
@@ -914,11 +924,6 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
num_experts = model_config.moe_num_experts
num_experts_per_rank = num_experts // parallel_config.expert_parallel_size
num_experts_start_offset = expert_parallel_rank * num_experts_per_rank
max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
parallel_config.local_data_parallel_id = parallel_config.data_parallel_rank % (
max_chips_per_node // parallel_config.tensor_parallel_size
)
parallel_config.expert_parallel_rank = expert_parallel_rank
parallel_config.num_experts_per_rank = num_experts_per_rank
parallel_config.num_experts_start_offset = num_experts_start_offset