Skip to content

notebooks: guard nvidia-smi for CPU-only runtimes (G1.2) #11

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mattdani21 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

notebooks: guard nvidia-smi for CPU-only runtimes (G1.2) #11

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 21 additions & 5 deletions notebooks/colab-phase1-eval.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,19 @@
"CATEGORIES = 'bugfix,feature' # or 'refactor'\n",
"import os, json, subprocess, sys, time, glob, shutil, urllib.request\n",
"print('python', sys.version.split()[0], '| scope:', SCOPE, '| categories:', CATEGORIES)\n",
"print(subprocess.run(['nvidia-smi'], capture_output=True, text=True).stdout[:900])\n",
"gpu = subprocess.run(['nvidia-smi', '--query-gpu=memory.total,name', '--format=csv,noheader'], capture_output=True, text=True).stdout.strip().splitlines()\n",
"# nvidia-smi is absent on CPU-only runtimes (no NVIDIA driver). Guard so the\n",
"# cell reports the situation instead of crashing with FileNotFoundError.\n",
"def nvidia_smi(*args):\n",
" try:\n",
" return subprocess.run(['nvidia-smi', *args], capture_output=True, text=True, timeout=30).stdout\n",
" except (FileNotFoundError, subprocess.TimeoutExpired):\n",
" return ''\n",
"print((nvidia_smi() or 'nvidia-smi NOT FOUND (CPU-only runtime — no NVIDIA driver).').strip()[:900])\n",
"gpu = nvidia_smi('--query-gpu=memory.total,name', '--format=csv,noheader').strip().splitlines()\n",
"vram = sum(int(line.split()[0].replace(',', '')) for line in gpu if line) if gpu else 0\n",
"print('GPU count:', len(gpu), '| total VRAM MiB:', vram)\n",
"if not gpu:\n",
" print('WARNING: no GPU detected — eval will run on CPU (slow). Set Runtime -> Change runtime type -> Hardware accelerator: T4 GPU')\n",
"if SCOPE == 'full':\n",
" assert vram >= 20000, 'full scope needs >=20GB VRAM (L4/A100) — T4 15GB cannot hold 27B Q4; change runtime type'\n",
"elif vram < 8000:\n",
Expand Down Expand Up @@ -218,9 +227,16 @@
}
],
"metadata": {
"kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"},
"language_info": {"name": "python", "version": "3.11"}
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"name": "python",
"version": "3.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
}
13 changes: 9 additions & 4 deletions notebooks/phase1-pipeline-eval.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,12 @@
"import os, json, subprocess, sys, time, glob, shutil, urllib.request\n",
"CATEGORIES = 'bugfix,feature'\n",
"print('python', sys.version.split()[0], '| categories:', CATEGORIES)\n",
"print(subprocess.run(['nvidia-smi'], capture_output=True, text=True).stdout[:800])\n"
"def nvidia_smi(*args):\n",
" try:\n",
" return subprocess.run(['nvidia-smi', *args], capture_output=True, text=True, timeout=30).stdout\n",
" except (FileNotFoundError, subprocess.TimeoutExpired):\n",
" return ''\n",
"print((nvidia_smi() or 'nvidia-smi NOT FOUND (CPU-only runtime — no NVIDIA driver).').strip()[:800])\n"
]
},
{
Expand Down Expand Up @@ -74,7 +79,7 @@
"outputs": [],
"source": [
"# CUDA layout on Kaggle (diag-verified): toolkit 12.8, driver lib ONLY in\n",
"# /usr/local/cuda-12.8/compat \u2014 symlink into toolkit lib64 for CMake.\n",
"# /usr/local/cuda-12.8/compat — symlink into toolkit lib64 for CMake.\n",
"COMPAT = '/usr/local/cuda-12.8/compat'\n",
"for lib in ['libcuda.so', 'libcuda.so.1']:\n",
" src = COMPAT + '/' + lib\n",
Expand Down Expand Up @@ -103,7 +108,7 @@
"bin_path = '/kaggle/working/llama.cpp/build/bin/llama-server'\n",
"assert os.path.exists(bin_path), 'llama-server build failed'\n",
"# Trim build footprint but KEEP the shared libs: llama-server links\n",
"# libggml-cuda.so etc from build/bin \u2014 copying only the binary broke exec\n",
"# libggml-cuda.so etc from build/bin — copying only the binary broke exec\n",
"# (rc=127, missing libs) on run 9. Copy the whole bin dir + ldd sanity.\n",
"subprocess.run(['cp', '-r', '/kaggle/working/llama.cpp/build/bin', '/kaggle/working/llama-bin'], check=True)\n",
"subprocess.run(['rm', '-rf', '/kaggle/working/llama.cpp'])\n",
Expand Down Expand Up @@ -141,7 +146,7 @@
" print((r.stdout or '')[-2500:])\n",
" print((r.stderr or '')[-1000:])\n",
"except subprocess.TimeoutExpired:\n",
" print('PIPELINE TIMEOUT after', round((time.time() - t0) / 60, 1), 'min \u2014 partial results kept')\n",
" print('PIPELINE TIMEOUT after', round((time.time() - t0) / 60, 1), 'min — partial results kept')\n",
"# Keep the kernel output small: the models/ dir (fallback downloads) would\n",
"# bloat /kaggle/working into the output snapshot and break result fetches.\n",
"shutil.rmtree('/kaggle/working/models', ignore_errors=True)\n",
Expand Down
Loading