scripts/install_rtx6000ada.sh

#!/usr/bin/env bash
set -euo pipefail

# Install CUDA-enabled PyTorch and the training stack for RTX 6000 Ada.
# Run this instead of plain `pip install -r requirements.txt` if your server accidentally installed CPU-only torch.

python -m pip install -U pip setuptools wheel

# CUDA 12.4 PyTorch wheels work on recent NVIDIA drivers. If your driver only supports older CUDA,
# use the matching PyTorch index from https://pytorch.org/get-started/locally/ .
python -m pip install --upgrade --index-url https://download.pytorch.org/whl/cu124 \
  torch torchvision torchaudio

# Install the rest. The CUDA torch wheel above already satisfies torch>=2.6, so pip should not replace it.
python -m pip install --upgrade \
  transformers 'trl[peft]' peft accelerate datasets bitsandbytes safetensors huggingface_hub trackio \
  pandas numpy tqdm jsonschema scikit-learn pyyaml rich

python - <<'PY'
import sys
import torch
print('python', sys.version)
print('torch', torch.__version__)
print('torch.version.cuda', torch.version.cuda)
print('cuda_available', torch.cuda.is_available())
if not torch.cuda.is_available():
    raise SystemExit('ERROR: CUDA is not available. Check NVIDIA driver, CUDA_VISIBLE_DEVICES, and PyTorch CUDA wheel.')
print('gpu_count', torch.cuda.device_count())
print('gpu_name', torch.cuda.get_device_name(0))
PY