lupine 0.2.0

Small PyTorch adapter helpers for LUPINE-backed CUDA devices

lupine Python adapter

This package provides small PyTorch helpers for LUPINE. It intentionally returns ordinary torch.device("cuda:N") objects so PyTorch continues to use its normal CUDA dispatch path while LUPINE handles CUDA driver/NVML calls underneath.

Declare all LUPINE hosts before any PyTorch CUDA work:

import lupine

with lupine.connect(host="<server>:14833") as s:
    device = s.device()
    model = model.to(device)

connect() loads the LUPINE libcuda.so.1 from ../build/libcuda.so.1 when used from this repository. For an installed package, pass libcuda=... or set LUPINE_LIBCUDA if the library lives somewhere else:

with lupine.connect(host="<server>:14833", libcuda="/opt/lupine/libcuda.so.1") as s:
    device = s.device()

For multiple LUPINE servers, pass the full host list in one call. The order defines the CUDA ordinals that PyTorch sees:

import lupine

with lupine.connect(host=["<server-a>:14833", "<server-b>:14833"]) as s:
    gpu0, gpu1 = s.devices()
    model0 = model0.to(gpu0)  # cuda:0
    model1 = model1.to(gpu1)  # cuda:1

Do not add a second host after tensors have already been moved to the first one. LUPINE opens connections from LUPINE_SERVER when CUDA first initializes, and later changes to LUPINE_SERVER are not picked up by the current process.

Exiting the context restores LUPINE_SERVER only if CUDA was not initialized inside the block. If CUDA was initialized, the process-global LUPINE connection is already active and cannot be disconnected safely.

The adapter does not create a new PyTorch backend such as torch.device("lupine"). A true custom PyTorch device would require registering PrivateUse1 kernels and backend support. LUPINE already works best when PyTorch sees CUDA tensors and the LUPINE library is selected through the dynamic linker.