[shardformer] support ep for deepseek v3 (#6185)

* [feature] support ep for deepseek v3

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix test

* [shardformer] fix deepseek v3 init

* [lazy] fit lora for lazy init

* [example] support npu for deepseek v3

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
Hongxin Liu
2025-02-11 16:10:25 +08:00
committed by GitHub
parent 17062c83b9
commit 2b415e5999
13 changed files with 612 additions and 22 deletions

View File

@@ -64,7 +64,10 @@ class ProcessGroupMesh:
system resources.
"""
for group in self._ranks_to_group.values():
dist.destroy_process_group(group)
try:
dist.destroy_process_group(group)
except ValueError:
pass
# Manually clear all process groups to save memory
gc.collect()