Instructions to use krystv/nomen-ai with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use krystv/nomen-ai with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Add GPU preflight diagnostics
Browse files- scripts/preflight_gpu.py +27 -0
scripts/preflight_gpu.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""GPU preflight diagnostics for Nomen-AI training.
|
| 2 |
+
|
| 3 |
+
Run before smoke/SFT/DPO to fail early if CUDA or VRAM is unavailable.
|
| 4 |
+
"""
|
| 5 |
+
import sys
|
| 6 |
+
import torch
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
def main():
|
| 10 |
+
print('torch:', torch.__version__)
|
| 11 |
+
print('cuda_available:', torch.cuda.is_available())
|
| 12 |
+
if not torch.cuda.is_available():
|
| 13 |
+
raise SystemExit('ERROR: CUDA is not available. Use a Colab GPU/T4 runtime or Docker with NVIDIA runtime.')
|
| 14 |
+
device = torch.cuda.current_device()
|
| 15 |
+
name = torch.cuda.get_device_name(device)
|
| 16 |
+
props = torch.cuda.get_device_properties(device)
|
| 17 |
+
total_gb = props.total_memory / 1e9
|
| 18 |
+
print('gpu_name:', name)
|
| 19 |
+
print('total_vram_gb:', round(total_gb, 2))
|
| 20 |
+
print('compute_capability:', f'{props.major}.{props.minor}')
|
| 21 |
+
if total_gb < 14:
|
| 22 |
+
raise SystemExit(f'ERROR: VRAM {total_gb:.1f}GB is below expected T4-class 15GB. Use T4/A10G or larger.')
|
| 23 |
+
print('GPU_PREFLIGHT_PASS')
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
if __name__ == '__main__':
|
| 27 |
+
main()
|