code/slurm_logs/build_and_train_bird_pairwise.sbatch

#!/bin/bash
#SBATCH --job-name=vl
#SBATCH --partition=gpu-large
#SBATCH --qos=batch-long
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=8
#SBATCH --mem=120G
#SBATCH --time=10:00:00
#SBATCH --output=/weka/s225250685/mats-tist/slurm_logs/build_and_train_bird_pairwise_%j.out

# Build BIRD-train pairwise dataset (after K=30 gen completes) + SFT Llama-3.2-3B.
# Chained after the K=30 gen job (89630).

set -u
cd /weka/s225250685/mats-tist
set -a; source /weka/s225250685/mats-tist/.env; set +a
export HF_HOME=/weka/s225250685/Huggingface HF_HUB_CACHE=/weka/s225250685/Huggingface/hub
export PYTHONNOUSERSITE=1 NO_PROXY=localhost,127.0.0.1
export PYTHONPATH=/weka/s225250685/mats-tist TOKENIZERS_PARALLELISM=false
export DB_EXEC_API_DISABLE=1

PY=/weka/s225250685/conda-envs/handbook/bin/python

CANDIDATES_FILE=data/qwen72b_candidates_bird_train_K30.jsonl
DATASET_OUT=data/sft_pairwise_bird_train_K30_v2
SFT_OUT=alignment-handbook/output/selector-llama3b-pairwise-bird-K30

echo "[$(date)] Step 1: build BIRD pairwise dataset"
$PY scripts/build_pairwise_sft_v2.py \
    --source bird_train \
    --input "$CANDIDATES_FILE" \
    --out "$DATASET_OUT" \
    --max_yn 6 --max_nn 1 \
    2>&1

echo "[$(date)] Step 2: SFT Llama-3.2-3B on BIRD pairwise"
$PY scripts/train_selector_pairwise_thin.py \
    --base meta-llama/Llama-3.2-3B-Instruct \
    --data "$DATASET_OUT" \
    --out "$SFT_OUT" \
    --epochs 2 --lr 2e-5 --bs 8 --grad_accum 16 --max_len 4096 \
    2>&1

echo "[$(date)] DONE"