Stop wasting money on GPU runs that fail silently

ComputeFence catches environment, storage, and configuration mistakes before your training run starts. 30 seconds. Free and open source.

ComputeFence v0.3.1 — Pre-flight diagnostic
━━━━━━━━━━━━━━━━━━━━
2 WARNINGS · 0 BLOCKERS · 4 PASSEDEnvironment
✓ PyTorch 2.4.0 detected
✓ CUDA available — NVIDIA A40 (1 GPU)
✓ Accelerate config matches GPU countStorage
⚠ HFHOME is not set — model cache will use ephemeral disk
Fix: export HFHOME=/workspace/.cache/huggingface
⚠ Checkpoint output directory is not on a persistent volume
Fix: point --output-dir at an attached network volume
✓ Disk headroom: 82 GB free━━━━━━━━━━━━━━━━━━━━
2 warning(s) found. Review before launching.
Exit codes: 0 = all passed, 1 = warnings, 2 = blockers

What it checks — replace the bullet points with these:

CUDA and GPU visibility — confirms your GPU is accessible and training will not silently fall back to CPU.

HuggingFace cache path — confirms model files write to persistent storage not ephemeral disk.


Accelerate GPU count — confirms your config matches the GPUs actually on the instance.


Dataset integrity — optional scan for duplicates and conflicting labels.

• Get release updates - Click Here

Already available now — pip install computefence