Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →nvidia/test-writing
Guide for writing tests in torch-harmonics.
Passed 21 of 21 test cases with the skill, and 4 of 21 without it.
and 27 more in the full report
18 fixed4 passed either way0 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 17 test cases the skill fixed. Same model, same prompt, with and without the skill.
Construct the class structure and initialization method for `TestLayerNorm` running on CUDA devices to ensure precise float32 matrix multiplications.
| Checked | Without skill | With skill |
|---|---|---|
| Calls disable_tf32 within the setUpClass method | Missed | Passed |
The instructions your model reads when the skill loads.
tests/ to pick up naming conventions, helper usage,
and parameterized.expand patterns.The files for serial tests should be named test_<feature_name>.py
and distributed tests test_distributed_<feature_name>.py respectively.
| dtype | atol | rtol |
|---|---|---|
| fp64 | 1e-6 | 1e-6 |
| fp32 | 1e-6 | 1e-5 |
| fp16 | 1e-3 | 1e-2 |
| bf16 | 1e-2 | 5e-2 |
Parameter gradients accumulate noise over batch × spatial dims, so loosen their tolerance by a factor of 10–1000× relative to per-element output tolerances (use judgment based on reduction size). For stacked/chained layers, loosen further (2–5×).
test cases passed, vs 4 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.