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

[XPU] ep+tp all2all (#4836)

Browse Source
This commit is contained in:
zhupengyang
2025-11-06 17:26:14 +08:00
committed by GitHub
parent 901d559aa7
commit b54eb7ad81
8 changed files with 201 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
fastdeploy/config.py
View File

@@ -577,6 +577,15 @@ class ParallelConfig:
else:
self.pd_disaggregation_mode = "None"
# ep+tp strategy: "all_reduce" or "all_to_all"
# all_reduce: qkv_linear + attn + out_linear + allreduce
# all_to_all: allgather + qkv_linear + attn + all2all + out_linear
self.ep_tp_strategy = envs.FD_EP_TP_STRATEGY
assert self.ep_tp_strategy in [
"all_reduce",
"all_to_all",
], f"FD_EP_TP_STRATEGY: '{self.ep_tp_strategy}' is not supported, only supports 'all_reduce' or 'all_to_all'."
def set_communicate_group(self):
# different tp group id
# prevent different tp_groups using the same group_id