For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Python class
PinnedDeviceTensor
PinnedDeviceTensor
max.experimental.nn.PinnedDeviceTensor
Type alias for a Tensor parameter that Module.to will leave on its current device.
Use this for parameters that must stay on a specific device regardless of where the rest of the module is moved. For example, scalar quantization scale factors that GPU kernels consume as host-side launch arguments should remain on CPU; moving them to the accelerator would force an expensive device sync every forward pass.