Operator Doc Template

Copy this template when adding a new operator page.

Summary

Describe what the operator does, which RL workload it targets, and why it exists.

Entry Point

# Show the public Python API used by callers.

Backends

Backend	Wrapper	Native symbol	Status
CUDA
ROCm
PyTorch fallback

Tensor Contract

Argument	Shape	Dtype	Requirements

Dispatch Behavior

Explain how the registry chooses this operator and what happens when a backend is unavailable.

Accuracy

Document the PyTorch reference implementation, tolerance, and any dtype-specific behavior.

Performance Notes

List the benchmark command and the workload sizes used to evaluate the operator.

Tests

python -m pytest path/to/test.py -v

Known Limitations

Track unsupported shapes, dtypes, devices, or architecture constraints.