Upload thunder_setup.sh
Browse files- thunder_setup.sh +76 -0
thunder_setup.sh
ADDED
|
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
# Thunder Compute H100 instance kurulum scripti
|
| 3 |
+
# Kullanim:
|
| 4 |
+
# chmod +x thunder_setup.sh
|
| 5 |
+
# ./thunder_setup.sh
|
| 6 |
+
# HF_TOKEN=<token> ./thunder_setup.sh musabc
|
| 7 |
+
|
| 8 |
+
set -e
|
| 9 |
+
|
| 10 |
+
USER="${1:-musabc}"
|
| 11 |
+
echo "=========================================="
|
| 12 |
+
echo "Thunder Compute V5 Bootstrap"
|
| 13 |
+
echo "HF user: $USER"
|
| 14 |
+
echo "=========================================="
|
| 15 |
+
|
| 16 |
+
# GPU kontrol
|
| 17 |
+
echo ""
|
| 18 |
+
echo "[1/6] GPU kontrol..."
|
| 19 |
+
nvidia-smi | head -25
|
| 20 |
+
GPU_NAME=$(nvidia-smi --query-gpu=name --format=csv,noheader | head -1)
|
| 21 |
+
echo " GPU: $GPU_NAME"
|
| 22 |
+
|
| 23 |
+
# Python ortam
|
| 24 |
+
echo ""
|
| 25 |
+
echo "[2/6] Python ortami..."
|
| 26 |
+
python --version
|
| 27 |
+
pip install --upgrade pip
|
| 28 |
+
|
| 29 |
+
# PyTorch (H100 icin CUDA 12.4+ gerek)
|
| 30 |
+
echo ""
|
| 31 |
+
echo "[3/6] PyTorch (CUDA 12.4) kurulumu..."
|
| 32 |
+
pip install --index-url https://download.pytorch.org/whl/cu124 \
|
| 33 |
+
torch torchvision torchaudio
|
| 34 |
+
|
| 35 |
+
# Diger bagimliliklar
|
| 36 |
+
echo ""
|
| 37 |
+
echo "[4/6] Bagimliliklar..."
|
| 38 |
+
pip install \
|
| 39 |
+
huggingface_hub \
|
| 40 |
+
tokenizers \
|
| 41 |
+
numpy \
|
| 42 |
+
tqdm \
|
| 43 |
+
pyarrow \
|
| 44 |
+
liger-kernel
|
| 45 |
+
|
| 46 |
+
# HF login kontrol
|
| 47 |
+
echo ""
|
| 48 |
+
echo "[5/6] HuggingFace token kontrol..."
|
| 49 |
+
if [ -z "$HF_TOKEN" ]; then
|
| 50 |
+
echo " HF_TOKEN env var yok — public repo varsayilir"
|
| 51 |
+
echo " Private icin: export HF_TOKEN=hf_..."
|
| 52 |
+
fi
|
| 53 |
+
|
| 54 |
+
# Veri + ckpt + kod indir
|
| 55 |
+
echo ""
|
| 56 |
+
echo "[6/6] Veri + kod indiriliyor..."
|
| 57 |
+
if [ ! -f "hf_pull_v5.py" ]; then
|
| 58 |
+
echo " hf_pull_v5.py yok, HF'ten cekiliyor..."
|
| 59 |
+
huggingface-cli download "$USER/nanogpt-tr-v5-code" \
|
| 60 |
+
hf_pull_v5.py --local-dir . || \
|
| 61 |
+
wget "https://huggingface.co/$USER/nanogpt-tr-v5-code/resolve/main/hf_pull_v5.py"
|
| 62 |
+
fi
|
| 63 |
+
python hf_pull_v5.py --user "$USER"
|
| 64 |
+
|
| 65 |
+
echo ""
|
| 66 |
+
echo "=========================================="
|
| 67 |
+
echo "KURULUM TAMAM"
|
| 68 |
+
echo "=========================================="
|
| 69 |
+
echo ""
|
| 70 |
+
echo "GPU: $GPU_NAME"
|
| 71 |
+
echo ""
|
| 72 |
+
echo "Egitime devam:"
|
| 73 |
+
echo " python 05_train_v5.py --compile --resume"
|
| 74 |
+
echo ""
|
| 75 |
+
echo "Sample testi:"
|
| 76 |
+
echo " python 06_sample.py --latest"
|