github
/
minimind
mirror of https://github.com/jingyaogong/minimind.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge 0c91dea4a2 into cc312c1cc6

Yuchen Lv 2026-09-19 09:03:44 +08:00 committed by GitHub
parent cc312c1cc6 0c91dea4a2
commit 87d708e4d2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 0 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
trainer/train_agent.py
View File

@ -471,7 +471,6 @@ if __name__ == "__main__":
if args.use_compile == 1: if args.use_compile == 1:
model = torch.compile(model) model = torch.compile(model)
Logger('torch.compile enabled') Logger('torch.compile enabled')
rollout_engine.update_policy(model)
if dist.is_initialized(): if dist.is_initialized():
# 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费 # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False) model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)

1
trainer/train_grpo.py
View File

@ -312,7 +312,6 @@ if __name__ == "__main__":
if args.use_compile == 1: if args.use_compile == 1:
model = torch.compile(model) model = torch.compile(model)
Logger('torch.compile enabled') Logger('torch.compile enabled')
rollout_engine.update_policy(model)
if dist.is_initialized(): if dist.is_initialized():
# 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费 # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False) model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)

1
trainer/train_ppo.py
View File

@ -427,7 +427,6 @@ if __name__ == "__main__":
if args.use_compile == 1: if args.use_compile == 1:
actor_model = torch.compile(actor_model) actor_model = torch.compile(actor_model)
Logger('torch.compile enabled') Logger('torch.compile enabled')
rollout_engine.update_policy(actor_model)
if dist.is_initialized(): if dist.is_initialized():
# freqs_cos/freqs_sin 各 rank 由 config 确定性算出，默认每步广播一次纯属浪费 # freqs_cos/freqs_sin 各 rank 由 config 确定性算出，默认每步广播一次纯属浪费
actor_model = DistributedDataParallel(actor_model, device_ids=[local_rank], broadcast_buffers=False) actor_model = DistributedDataParallel(actor_model, device_ids=[local_rank], broadcast_buffers=False)