reference/generated/torchrl.modules.get_recurrent_matmul_precision

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

get_recurrent_matmul_precision

torchrl.modules.get_recurrent_matmul_precision() → Literal['ieee', 'tf32', 'tf32x3']

Resolve the currently effective precision to a concrete mode.

Always returns one of "ieee", "tf32" or "tf32x3". The result is what the kernel actually runs at, including preset / GPU resolution. Does not see per-module overrides; those are resolved at the call site via _resolve_precision().