1 files changed, 29 insertions, 0 deletions
diff --git a/collaborativeagents/training/train_sft_qlora.sbatch b/collaborativeagents/training/train_sft_qlora.sbatch
new file mode 100644
index 0000000..81c3db1
--- /dev/null
+++ b/collaborativeagents/training/train_sft_qlora.sbatch
@@ -0,0 +1,29 @@
+#!/bin/bash
+#SBATCH --job-name=sft_qlora
+#SBATCH --output=sft_qlora_%j.out
+#SBATCH --error=sft_qlora_%j.err
+#SBATCH --time=12:00:00
+#SBATCH --partition=a100
+#SBATCH --gres=gpu:1
+#SBATCH --cpus-per-task=8
+#SBATCH --mem=64G
+
+# QLoRA SFT Training - Ultra Minimal Hardware (1 GPU, ~12GB VRAM)
+# Can run on A100 40GB, V100, or even consumer GPUs
+
+cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/training
+
+source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
+conda activate eval
+
+echo "=== Starting QLoRA SFT Training ==="
+echo "GPU: $(nvidia-smi --query-gpu=name --format=csv,noheader | head -1)"
+echo "Config: llama_factory_qlora_config.yaml"
+echo "Start time: $(date)"
+
+# Run training with LLaMA-Factory
+llamafactory-cli train llama_factory_qlora_config.yaml
+
+echo "=== Training Complete ==="
+echo "End time: $(date)"
+echo "Output: outputs/sft_reflection_qlora"