#!/bin/bash set -e echo "=== SIPA Binary SFT Pipeline ===" # 1. venv python3 -m venv ~/venv && source ~/venv/bin/activate # 2. PyTorch CUDA 12.4 pip install torch==2.5.1 --index-url https://download.pytorch.org/whl/cu124 --quiet # 3. Dependencies pip install transformers==4.46.3 accelerate==0.34.2 datasets==2.21.0 bitsandbytes==0.44.1 trl==0.12.2 peft==0.13.2 --quiet # 4. Verify CUDA python3 -c "import torch; print(f'CUDA OK: {torch.cuda.is_available()}')" # 5. Download dataset pip install huggingface_hub --quiet python3 -c " from huggingface_hub import hf_hub_download import shutil path = hf_hub_download('SoulInPsyAbstract/sipa-os-governance', 'AI_EXPERIMENTS/DATASETS/protocol0_binary_sft.jsonl', repo_type='dataset', token='$HF_TOKEN') shutil.copy(path, '$HOME/protocol0_binary_sft.jsonl') print('Dataset OK') " # 6. Download training scripts for script in train_hermes3_std.py train_qwen25_std.py train_r1_std.py; do curl -fsSL "https://raw.githubusercontent.com/soulinpsyabstract/sipa-os-governance/main/scripts/$script" -o "$HOME/$script" done # 7. Train source ~/venv/bin/activate python3 ~/train_hermes3_std.py 2>&1 | tee ~/train_hermes3.log echo "=== HERMES DONE ===" python3 ~/train_qwen25_std.py 2>&1 | tee ~/train_qwen25.log echo "=== QWEN DONE ===" python3 ~/train_r1_std.py 2>&1 | tee ~/train_r1.log echo "=== ALL THREE DONE ==="