[PyTorch] [torch.compile] torch.compile support for Linear#3053

Draft

pggPL wants to merge 20 commits into

NVIDIA:mainfrom

pggPL:linear_torch_compile_final_attempt

Commits on May 12, 2026

Commits on May 13, 2026

[PyTorch] Iterate on torch.compile support for Linear
pggPL
committed

Commits on May 14, 2026

[PyTorch] Fix indentation errors in dynamo.py
pggPL
committed

Commits on May 27, 2026

[PyTorch] Replace Linear fake-impls with output-info descriptors
pggPL
committed

Commits on May 28, 2026

[PyTorch] Unify TensorSpec hierarchy and infer num_outputs dynamically
pggPL
committed
[PyTorch] Require output-info descriptors for custom ops
pggPL
committed
[PyTorch] Drop dead parameters from torch.compile custom op registration
pggPL
committed
[PyTorch] Unify quantized tensor flatten via declarative schema
pggPL
committed
[PyTorch] Generic Quantizer.create_storage_metadata via declarative schema
pggPL
committed
[PyTorch] Generic Quantizer._flatten / _do_unflatten via declarative schema
pggPL
committed
[PyTorch] Drop dead code and stale comments from torch.compile branch
pggPL
committed
[PyTorch] Generic Quantizer.create_metadata via declarative schema
pggPL
committed
[PyTorch] Drop Recipe._flatten/_unflatten torch.compile protocol
pggPL
committed
[PyTorch] Consolidate tex.DType opaque-type registration in fp8_dtype.py
pggPL
committed

Commits on May 29, 2026

[PyTorch] Drop TensorSpec; reassemble torch.compile outputs from fake templates
pggPL
committed