torch.tvarant¶

Importing torch_tvarant registers the tvarant PrivateUse1 backend and exposes helpers under torch.tvarant (same idea as torch.cuda / torch.mps).

import torch
import torch_tvarant

assert torch.tvarant.is_available()
x = torch.empty(2, 3, device="tvarant")

Click an API below for the full signature, parameters, and examples.

Device¶

API	Summary
`is_available`	Whether the backend can run
`is_initialized`	Lazy init completed
`device_count`	Number of logical devices
`current_device`	Current device index
`set_device`	Select device index
`synchronize`	Wait for kernels
`backend`	`"sim"` or `"opencl"`

RNG¶

API	Summary
`manual_seed`	Seed Tvarant RNG
`manual_seed_all`	Seed all devices
`get_rng_state`	Read RNG state
`set_rng_state`	Restore RNG state