Rebrand the project as "ComfyUI Quantization Toolkit" and introduce native INT4 (ConvRot W4A4) support, mixed INT4/INT8 modes, and expanded quantization policies. Add an INT4 torch.compile compatibility shim, INT4-aware lazy-compile changes, richer LoRA stack nodes with autogrow, and a Quantized LoRA entry/apply pair. Replace ad-hoc prints with structured logging, bump package metadata to 2.0.0, and add documentation (advanced usage, checkpoints, policies) plus many unit tests for compile, lazy-compile, LoRA, adapter caching, saving, and modes. Misc: kernel config, model save/load, and numerous runtime/adapter refactors and robustness fixes.
Add a new INT8 Lazy Torch Compile node and register it in the package; update README to document lazy compile, runtime backends, LoRA order, and usage guidance. Improve Triton fused kernels with safer tail handling, optional legacy-unsafe mode, support for prepacked weight layouts, and contiguous input handling; expose HAS_MN_TAIL / LEGACY_UNSAFE kernel flags and adjust launch parameters. Add runtime/backend controls to the INT8 model adapter (runtime_backend, small_batch_fallback, prepack_int8_weights), persist prepacked weight buffer on quantized modules, print/reset runtime stats, and wire those options through adapter state and UI defaults. Add stable dynamic LoRA UUIDs for runtime tracking. Misc: update __init__ mappings, example workflow asset, and various CLI/README clarifications to reflect new defaults and features.
ThereforeGames