Upload setup_a40.sh with huggingface_hub
Browse files- setup_a40.sh +75 -0
setup_a40.sh
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 3 |
+
# MonSub v3 โ A40 Setup Script
|
| 4 |
+
# RunPod A40 48GB pod ะดััั ะฐะถะธะปะปััะปะฝะฐ
|
| 5 |
+
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 6 |
+
|
| 7 |
+
set -e
|
| 8 |
+
|
| 9 |
+
# โโ HF Token โโ
|
| 10 |
+
export HF_TOKEN="${HF_TOKEN}" # RunPod env-ะด ัะพั
ะธัััะปะฝะฐ
|
| 11 |
+
export HUGGINGFACE_HUB_TOKEN="$HF_TOKEN"
|
| 12 |
+
|
| 13 |
+
# โโ ะงะฃะฅะะ: Cache-ะณ workspace volume ััั ัะธะณะปาฏาฏะปัั
โโ
|
| 14 |
+
# Container disk (50GB) ะดาฏาฏัะดัะณ โ workspace volume ะฐัะธะณะปะฐะฝะฐ
|
| 15 |
+
export HF_HOME=/workspace/.cache
|
| 16 |
+
export TMPDIR=/workspace/tmp
|
| 17 |
+
mkdir -p /workspace/.cache /workspace/tmp
|
| 18 |
+
|
| 19 |
+
echo "=============================================="
|
| 20 |
+
echo "MonSub v3 โ A40 Setup"
|
| 21 |
+
echo "HF_HOME=$HF_HOME"
|
| 22 |
+
echo "TMPDIR=$TMPDIR"
|
| 23 |
+
echo "=============================================="
|
| 24 |
+
|
| 25 |
+
# โโ Dependencies โโ
|
| 26 |
+
echo ""
|
| 27 |
+
echo "=== Installing dependencies ==="
|
| 28 |
+
pip install -q \
|
| 29 |
+
"transformers>=4.46.0" \
|
| 30 |
+
"datasets==2.21.0" \
|
| 31 |
+
accelerate \
|
| 32 |
+
evaluate \
|
| 33 |
+
jiwer \
|
| 34 |
+
soundfile \
|
| 35 |
+
librosa
|
| 36 |
+
|
| 37 |
+
# datasets==2.21.0 ะะะะะะ (latest โ torchcodec ImportError)
|
| 38 |
+
|
| 39 |
+
# โโ GPU check โโ
|
| 40 |
+
echo ""
|
| 41 |
+
echo "=== GPU Info ==="
|
| 42 |
+
python -c "
|
| 43 |
+
import torch
|
| 44 |
+
if torch.cuda.is_available():
|
| 45 |
+
name = torch.cuda.get_device_name(0)
|
| 46 |
+
vram = torch.cuda.get_device_properties(0).total_memory / 1e9
|
| 47 |
+
print(f'GPU: {name}')
|
| 48 |
+
print(f'VRAM: {vram:.1f}GB')
|
| 49 |
+
else:
|
| 50 |
+
print('WARNING: No GPU!')
|
| 51 |
+
"
|
| 52 |
+
|
| 53 |
+
# โโ Download training script โโ
|
| 54 |
+
echo ""
|
| 55 |
+
echo "=== Downloading training script ==="
|
| 56 |
+
cd /workspace
|
| 57 |
+
|
| 58 |
+
# HuggingFace-ััั ะฐะฒะฐั
(ััะฒัะป paste ั
ะธะนะฝั)
|
| 59 |
+
python -c "
|
| 60 |
+
from huggingface_hub import hf_hub_download
|
| 61 |
+
try:
|
| 62 |
+
path = hf_hub_download('Tsedee/monsub-training-scripts', 'run_finetune_v3.py', token=os.environ['HF_TOKEN'])
|
| 63 |
+
import shutil
|
| 64 |
+
shutil.copy(path, '/workspace/run_finetune_v3.py')
|
| 65 |
+
print('Downloaded from HF')
|
| 66 |
+
except:
|
| 67 |
+
print('HF download failed - paste the script manually')
|
| 68 |
+
"
|
| 69 |
+
|
| 70 |
+
# โโ Start training โโ
|
| 71 |
+
echo ""
|
| 72 |
+
echo "=== Starting v3 training ==="
|
| 73 |
+
echo "Log: /workspace/train_v3.log"
|
| 74 |
+
echo ""
|
| 75 |
+
python /workspace/run_finetune_v3.py 2>&1 | tee /workspace/train_v3.log
|