Skip to content

Update Llama check to use module instead of model #734

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lekurile merged 2 commits into from
Sep 14, 2023
Merged

Update Llama check to use module instead of model #734

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ba1280d
Update Llama check to use module instead of model
lekurile Sep 14, 2023
c4b8a5c
Update print_throughput_step3 function as well
lekurile Sep 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -540,7 +540,7 @@ def main():
print_rank_0(
f'Epoch: {epoch} | Step: {step} | PPO Epoch: {ppo_ep+1} | Actor Loss: {actor_loss_sum/inner_iter} | Critic Loss: {critic_loss_sum/inner_iter} | Unsupervised Loss: {unsup_loss_sum/inner_iter}',
args.global_rank)
print_throughput_step3(rlhf_engine.actor.model,
print_throughput_step3(rlhf_engine.actor.module,
rlhf_engine.critic, args, e2e_time,
trainer.generate_time, training_time,
args.global_rank)
Expand Down
2 changes: 1 addition & 1 deletion applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,7 @@ def _generate_sequence(self, prompts, mask, step):
# This has been added due to a probability/nan error that happens after
# meta-llama/Llama-2-7b-hf enabled do_sample:
# https://huggingface.co/meta-llama/Llama-2-7b-hf/commit/6fdf2e60f86ff2481f2241aaee459f85b5b0bbb9
if self.actor_model.model.config.model_type == "llama":
if self.actor_model.module.config.model_type == "llama":
kwargs = dict(do_sample=False)
else:
kwargs = dict()
Expand Down