Skip to content

athene reward online dpo fix, speed up ray init #1322

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

Open
uralik wants to merge 1 commit into
base: online_training
Choose a base branch
from
Open

athene reward online dpo fix, speed up ray init #1322

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
13 changes: 8 additions & 5 deletions src/fairseq2/recipes/lm/_online_finetune/_recipe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,3 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
Expand Down Expand Up @@ -252,11 +252,14 @@

vocab_size = tokenizer.vocab_info.size

# initialize ray and vllm actors
ray.init(
address=f"ray://{config.vllm.ray_cluster_ip_address}:10001",
namespace="vllm_workers",
)
if gangs.root.rank == 0:
# initialize ray and vllm actors
ray.init(
address=f"ray://{config.vllm.ray_cluster_ip_address}:10001",
namespace="vllm_workers",
)

gangs.root.barrier()

vllm_actors = {}
# go over actor configs and initialize all of them
Expand Down
7 changes: 7 additions & 0 deletions src/fairseq2/recipes/lm/_online_finetune/_rewards.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -477,6 +477,13 @@ def prepare_preference_batch(
reference_score_rejected=None,
)

prompt_lengths = [
l
for idx, l in enumerate(prompt_batch.prompt_lengths)
if idx not in dummy_batch_ids
]
reward_output["prompt_lengths"] = prompt_lengths

return batch, is_bad_batch, reward_output


Expand Down
Loading