sft.slurm

#!/bin/bash
#SBATCH --job-name=apigen-fine-tune
#SBATCH --partition=hopper-prod
#SBATCH --qos=normal
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --gpus-per-node=8
#SBATCH --output=./logs/%x-%j.out
#SBATCH --err=./logs/%x-%j.err
#SBATCH --time=02-00:00:00

set -ex

module load cuda/12.1

source .venv/bin/activate

srun --nodes=1 --ntasks=1 --export=ALL,ACCELERATE_LOG_LEVEL=info accelerate launch --config_file examples/accelerate_configs/deepspeed_zero3.yaml examples/scripts/sft.py \
    --run_name=Llama-3.2-1B-Instruct-APIGen-FC-v0.1 \
    --model_name_or_path="meta-llama/Llama-3.2-1B-Instruct" \
    --dataset_name="plaguss/apigen-synth-trl" \
    --report_to="wandb" \
    --learning_rate=5.0e-06 \
    --lr_scheduler_type="cosine" \
    --per_device_train_batch_size=6 \
    --per_device_eval_batch_size=6 \
    --do_eval \
    --eval_strategy="steps" \
    --gradient_accumulation_steps=2 \
    --output_dir="data/Llama-3.2-1B-Instruct-APIGen-FC-v0.1" \
    --logging_steps=5 \
    --eval_steps=50 \
    --num_train_epochs=2 \
    --max_steps=-1 \
    --warmup_steps=50 \
    --max_seq_length=2048 \
    --push_to_hub \
    --gradient_checkpointing \
    --bf16