Skip to content
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.

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

SFT configs for Qwen coder models #438

Draft
wants to merge 12 commits into
base: main
Choose a base branch
from
Draft

Conversation

edbeeching
Copy link
Collaborator

@edbeeching edbeeching commented Feb 26, 2025

Saving for reference, no need to review:

LR scan

sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.00 zero3 "--learning_rate 5e-6 --hub_model_revision v00.02 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.02"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.00 zero3 "--learning_rate 1e-5 --hub_model_revision v00.03 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.03"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.00 zero3 "--learning_rate 2e-5 --hub_model_revision v00.04 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.04"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.01 zero3 "--learning_rate 5e-6 --hub_model_revision v00.05 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.05"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.01 zero3 "--learning_rate 1e-5 --hub_model_revision v00.06 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.06"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.01 zero3 "--learning_rate 2e-5 --hub_model_revision v00.07 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.07"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v00.01 zero3 "--learning_rate 4e-5 --hub_model_revision v00.08 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v00.08"

sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.00 zero3 "--learning_rate 5e-6 --hub_model_revision v01.02 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.02"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.00 zero3 "--learning_rate 1e-5 --hub_model_revision v01.03 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.03"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.00 zero3 "--learning_rate 2e-5 --hub_model_revision v01.04 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.04"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.01 zero3 "--learning_rate 5e-6 --hub_model_revision v01.05 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.05"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.01 zero3 "--learning_rate 1e-5 --hub_model_revision v01.06 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.06"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.01 zero3 "--learning_rate 2e-5 --hub_model_revision v01.07 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.07"
sbatch --job-name or1_sft_Qwen2.5-Coder-7B --nodes 1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v01.01 zero3 "--learning_rate 4e-5 --hub_model_revision v00.08 --output_dir data/open-r1/Qwen2.5-Coder-7B-Instruct-SFT-v01.08"

@lewtun
Copy link
Member

lewtun commented Mar 3, 2025

My scans

# V02.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.00 --run_name=qwen-7B_codeforces-cot_v02.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.01 --run_name=qwen-7B_codeforces-cot_v02.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.02 --run_name=qwen-7B_codeforces-cot_v02.02 --wandb_entity huggingface --wandb_project open-r1'
# V02.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.10 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=1.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.10 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.10 --run_name=qwen-7B_codeforces-cot_v02.10 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.11 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=2.1e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.11 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.11 --run_name=qwen-7B_codeforces-cot_v02.11 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v02.12 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v02.00 zero3 '--learning_rate=4.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v02.12 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v02.12 --run_name=qwen-7B_codeforces-cot_v02.12 --wandb_entity huggingface --wandb_project open-r1'

# V03.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v03.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v03.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v03.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v03.00 --run_name=qwen-7B_codeforces-cot_v03.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v03.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v03.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v03.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v03.01 --run_name=qwen-7B_codeforces-cot_v03.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v03.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v03.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v03.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v03.02 --run_name=qwen-7B_codeforces-cot_v03.02 --wandb_entity huggingface --wandb_project open-r1'

# V04.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v04.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v04.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v04.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v04.00 --run_name=qwen-7B_codeforces-cot_v04.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v04.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v04.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v04.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v04.01 --run_name=qwen-7B_codeforces-cot_v04.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v04.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v04.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v04.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v04.02 --run_name=qwen-7B_codeforces-cot_v04.02 --wandb_entity huggingface --wandb_project open-r1'

# V05.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v05.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v05.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v05.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v05.00 --run_name=qwen-7B_codeforces-cot_v05.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v05.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v05.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v05.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v05.01 --run_name=qwen-7B_codeforces-cot_v05.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v05.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v05.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v05.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v05.02 --run_name=qwen-7B_codeforces-cot_v05.02 --wandb_entity huggingface --wandb_project open-r1'

# V06.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.00 --run_name=qwen-7B_codeforces-cot_v06.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.01 --run_name=qwen-7B_codeforces-cot_v06.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.02 --run_name=qwen-7B_codeforces-cot_v06.02 --wandb_entity huggingface --wandb_project open-r1'
# V06.0X = lr scan, packing=false
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.10 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=1.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.10 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.10 --run_name=qwen-7B_codeforces-cot_v06.10 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.11 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=2.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.11 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.11 --run_name=qwen-7B_codeforces-cot_v06.11 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v06.12 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v06.00 zero3 '--learning_rate=4.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v06.12 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v06.12 --run_name=qwen-7B_codeforces-cot_v06.12 --wandb_entity huggingface --wandb_project open-r1'

# V07.0X = lr scan, packing=true
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.00 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=1.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.00 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.00 --run_name=qwen-7B_codeforces-cot_v07.00 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.01 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=2.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.01 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.01 --run_name=qwen-7B_codeforces-cot_v07.01 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.02 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=4.0e-5 --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.02 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.02 --run_name=qwen-7B_codeforces-cot_v07.02 --wandb_entity huggingface --wandb_project open-r1'
# V07.0X = lr scan, packing=false
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.10 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=1.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.10 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.10 --run_name=qwen-7B_codeforces-cot_v07.10 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.11 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=2.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.11 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.11 --run_name=qwen-7B_codeforces-cot_v07.11 --wandb_entity huggingface --wandb_project open-r1'
sbatch --mail-type=ALL [email protected]  --output=/fsx/h4/logs/%x-%j.out --err=/fsx/h4/logs/%x-%j.err --job-name=qwen-7B_codeforces-cot_v07.12 --nodes=1 slurm/train.slurm Qwen2.5-Coder-7B-Instruct sft v07.00 zero3 '--learning_rate=4.0e-5 --packing=false --hub_model_id=open-r1/Qwen2.5-Coder-7B-Instruct-SFT --hub_model_revision=v07.12 --output_dir=data/Qwen2.5-Coder-7B-Instruct-SFT-v07.12 --run_name=qwen-7B_codeforces-cot_v07.12 --wandb_entity huggingface --wandb_project open-r1'

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants