g2n 0.5.0

A compiler backend for PyTorch: pointwise fusion, buffer-reuse planning, and a persistent cross-run compile cache.

g2n

A compiler backend for PyTorch: pointwise fusion, buffer-reuse planning, and a persistent cross-run compile cache so repeat builds skip recompilation.

pip install g2n

import torch
import g2n

model = MyModule().eval()
compiled = g2n.compile(model)
# or register as a torch.compile backend:
compiled = torch.compile(model, backend="g2n")

A license key unlocks the Pro and Enterprise features (enhanced buffer planner, persistent cache, multi-accelerator routing, model-zoo configurations). See the documentation at https://g2n.dev/docs.

Custom kernels (Pro / Enterprise)

With a licensed tier, the g2n backend runs a real custom compile pass: it fuses LayerNorm (and a trailing GELU) into a Triton kernel via a torch.library custom op, then hands the rest of the graph to TorchInductor. See ARCHITECTURE.md. Correctness is covered by tests/test_layernorm.py; benchmark on your own GPU before quoting any speedup.