From dc801c07cf38b0c495686463e6ca6f871a64440e Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Tue, 27 Jan 2026 09:57:37 -0600 Subject: Add collaborativeagents module and update gitignore - Add collaborativeagents subproject with adapters, agents, and evaluation modules - Update .gitignore to exclude large binary files (.whl, .tar), wandb logs, and results Co-Authored-By: Claude Opus 4.5 --- collaborativeagents/training/train_sft.sbatch | 57 +++++++++++++++++++++++++++ 1 file changed, 57 insertions(+) create mode 100644 collaborativeagents/training/train_sft.sbatch (limited to 'collaborativeagents/training/train_sft.sbatch') diff --git a/collaborativeagents/training/train_sft.sbatch b/collaborativeagents/training/train_sft.sbatch new file mode 100644 index 0000000..a60894d --- /dev/null +++ b/collaborativeagents/training/train_sft.sbatch @@ -0,0 +1,57 @@ +#!/bin/bash +#SBATCH --job-name=sft_train +#SBATCH --account=bfqt-delta-gpu +#SBATCH --partition=gpuH200x8 +#SBATCH --nodes=1 +#SBATCH --ntasks=1 +#SBATCH --cpus-per-task=32 +#SBATCH --gres=gpu:4 +#SBATCH --mem=256G +#SBATCH --time=08:00:00 +#SBATCH --output=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/sft_train_%j.out +#SBATCH --error=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/sft_train_%j.err + +set -e +cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents +source /u/yurenh2/miniforge3/etc/profile.d/conda.sh +conda activate eval + +export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface +export PYTHONPATH="${PWD}:${PWD}/../src:${PYTHONPATH}" +export WANDB_MODE=offline + +echo "=== SFT Training (H200) ===" +date +nvidia-smi --query-gpu=index,name,memory.total --format=csv + +MODEL_PATH="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct" +DATA_PATH="training/training_data/sft_training_data.json" +OUTPUT_DIR="training/outputs/sft_reflection" + +echo "Model: $MODEL_PATH" +echo "Data: $DATA_PATH" +echo "Output: $OUTPUT_DIR" + +# Count training examples +echo "" +echo "Training data size: $(wc -c < $DATA_PATH) bytes" +python -c "import json; data = json.load(open('$DATA_PATH')); print(f'Training examples: {len(data)}')" + +echo "" +echo "Starting SFT training..." + +# Training with 4 GPUs using FSDP for memory efficiency +accelerate launch --num_processes=4 --mixed_precision=bf16 \ + training/train_sft.py \ + --model-path $MODEL_PATH \ + --data-path $DATA_PATH \ + --output-dir $OUTPUT_DIR \ + --num-epochs 4 \ + --learning-rate 1e-6 \ + --batch-size 1 \ + --gradient-accumulation 16 + +echo "" +echo "Training complete!" +date +ls -la $OUTPUT_DIR/ -- cgit v1.2.3