Skip to content

Add WIP code GRPO configs #593

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

Jump to bottom
Draft
wants to merge 8 commits into
base: main
Choose a base branch
from
Draft

Add WIP code GRPO configs #593

wants to merge 8 commits into from

Conversation

edbeeching
Copy link
Collaborator

@edbeeching edbeeching commented Apr 11, 2025 •
edited
Loading

DO NOT MERGE / REVIEW

Saving for reference.

sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.20 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=1.0e-06 --num_iterations=4 --hub_model_revision=v05.20 --output_dir=data/open-r1-code-GRPO-v05.20 --run_name=open-r1-code-GRPO-v05.20-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.21 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=2.0e-06 --num_iterations=4 --hub_model_revision=v05.21 --output_dir=data/open-r1-code-GRPO-v05.21 --run_name=open-r1-code-GRPO-v05.21-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.22 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=4.0e-06 --num_iterations=4 --hub_model_revision=v05.22 --output_dir=data/open-r1-code-GRPO-v05.22 --run_name=open-r1-code-GRPO-v05.22-lr-4.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.23 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=8.0e-06 --num_iterations=4 --hub_model_revision=v05.23 --output_dir=data/open-r1-code-GRPO-v05.23 --run_name=open-r1-code-GRPO-v05.23-lr-8.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.24 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=1.0e-05 --num_iterations=4 --hub_model_revision=v05.24 --output_dir=data/open-r1-code-GRPO-v05.24 --run_name=open-r1-code-GRPO-v05.24-lr-1.0e-05 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'

sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.25 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=1.0e-06 --num_iterations=4 --hub_model_revision=v05.25 --output_dir=data/open-r1-code-GRPO-v05.25 --run_name=open-r1-code-GRPO-v05.25-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.26 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=2.0e-06 --num_iterations=4 --hub_model_revision=v05.26 --output_dir=data/open-r1-code-GRPO-v05.26 --run_name=open-r1-code-GRPO-v05.26-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1'

sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.27 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=1.0e-06 --hub_model_revision=v05.27 --output_dir=data/open-r1-code-GRPO-v05.27 --run_name=open-r1-code-GRPO-v05.27-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1 --a2c_loss'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.28 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.20 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v05.28 --output_dir=data/open-r1-code-GRPO-v05.28 --run_name=open-r1-code-GRPO-v05.28-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1 --a2c_loss'



sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.30 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=1.0e-06 --num_iterations=4 --hub_model_revision=v05.30 --output_dir=data/open-r1-code-GRPO-v05.30 --run_name=open-r1-code-GRPO-v05.30-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.31 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=2.0e-06 --num_iterations=4 --hub_model_revision=v05.31 --output_dir=data/open-r1-code-GRPO-v05.31 --run_name=open-r1-code-GRPO-v05.31-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.32 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=4.0e-06 --num_iterations=4 --hub_model_revision=v05.32 --output_dir=data/open-r1-code-GRPO-v05.32 --run_name=open-r1-code-GRPO-v05.32-lr-4.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.33 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=8.0e-06 --num_iterations=4 --hub_model_revision=v05.33 --output_dir=data/open-r1-code-GRPO-v05.33 --run_name=open-r1-code-GRPO-v05.33-lr-8.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.34 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=1.0e-05 --num_iterations=4 --hub_model_revision=v05.34 --output_dir=data/open-r1-code-GRPO-v05.34 --run_name=open-r1-code-GRPO-v05.34-lr-1.0e-05 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO'

sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.35 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=1.0e-06 --hub_model_revision=v05.35 --output_dir=data/open-r1-code-GRPO-v05.35 --run_name=open-r1-code-GRPO-v05.35-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.36 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v05.36 --output_dir=data/open-r1-code-GRPO-v05.36 --run_name=open-r1-code-GRPO-v05.36-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1'

sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.37 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=1.0e-06 --hub_model_revision=v05.37 --output_dir=data/open-r1-code-GRPO-v05.37 --run_name=open-r1-code-GRPO-v05.37-lr-1.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1 --a2c_loss'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v05.38 --nodes=2 slurm/train.slurm Qwen2.5-7B-Instruct grpo v05.30 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v05.38 --output_dir=data/open-r1-code-GRPO-v05.38 --run_name=open-r1-code-GRPO-v05.38-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --num_iterations=1 --a2c_loss'

#  Latest R1 distill exps
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v6.00 --nodes=2 slurm/train.slurm DeepSeek-R1-Distill-Qwen-7B grpo v06.00 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v06.00 --output_dir=data/open-r1-code-GRPO-v06.00 --run_name=open-r1-code-GRPO-v06.00-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --gradient_accumulation_steps=16'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v6.01 --nodes=2 slurm/train.slurm DeepSeek-R1-Distill-Qwen-7B grpo v06.00 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v06.01 --output_dir=data/open-r1-code-GRPO-v06.01 --run_name=open-r1-code-GRPO-v06.01-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --gradient_accumulation_steps=32'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v6.02 --nodes=2 slurm/train.slurm DeepSeek-R1-Distill-Qwen-7B grpo v06.00 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v06.02 --output_dir=data/open-r1-code-GRPO-v06.02 --run_name=open-r1-code-GRPO-v06.02-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --gradient_accumulation_steps=64'
sbatch --mail-type=ALL [email protected]  --job-name=open-r1-code-GRPO-v6.03 --nodes=2 slurm/train.slurm DeepSeek-R1-Distill-Qwen-7B grpo v06.00 zero3 '--learning_rate=2.0e-06 --hub_model_revision=v06.03 --output_dir=data/open-r1-code-GRPO-v06.03 --run_name=open-r1-code-GRPO-v06.03-lr-2.0e-06 --wandb_entity=huggingface --wandb_project=open-r1 --wandb_run_group=open-r1-code-GRPO --gradient_accumulation_steps=128'

@edbeeching edbeeching requested a review from lewtun April 11, 2025 09:27
@lewtun lewtun mentioned this pull request May 2, 2025
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@lewtun lewtun Awaiting requested review from lewtun

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@edbeeching @lewtun