lmflow.utils.envs#

Functions#

`is_accelerate_env`()	Return True if any environment variable name starts with `ACCELERATE_`.
`require_cuda_for_gpu_mode`(→ None)	Raise if GPU execution was requested but CUDA is not available.
`set_cuda_device`(→ None)	Bind this process to `local_rank` on CUDA; raises if CUDA is unavailable.
`get_device_name`(→ str)	Get the device name based on the current machine.
`get_torch_device`(→ Any)	Return `torch.<device_name>` for the current device name.

lmflow.utils.envs.is_accelerate_env()[source]#: Return True if any environment variable name starts with ACCELERATE_.

lmflow.utils.envs.require_cuda_for_gpu_mode() → None[source]#: Raise if GPU execution was requested but CUDA is not available.

lmflow.utils.envs.set_cuda_device(local_rank: int) → None[source]#: Bind this process to local_rank on CUDA; raises if CUDA is unavailable.

lmflow.utils.envs.get_device_name() → str[source]#: Get the device name based on the current machine.

lmflow.utils.envs.get_torch_device() → Any[source]#

Return torch.<device_name> for the current device name.

If torch has no attribute with that name, logs a warning and returns torch.cuda as fallback.