Published by WaifuFinder. manifest.json pins 6 model files by SHA-256; the app verifies every one of them, plus the licence, attribution and validation files, before installation.

license: other library_name: coreml tags: - coreml - coremltools - apple-silicon - waifufinder - pgtformer - face-restoration license_name: PGTFormer BSD-style, non-commercial license_note: Redistribution is permitted for non-commercial use only. Commercial use requires permission from Xidian University or the contributors. See LICENSE and NOTICE.

Published by WaifuFinder. manifest.json pins 6 model files by SHA-256; the app verifies every one of them, plus the licence, attribution and validation files, before installation.

PGTFormer โ€” Swift/Core ML

Self-contained PGTFormer packages for WaifuFinder. WaifuFinder downloads these files directly: no Python, no source checkpoint, no conversion step.

Contents

Path What
PGTEncoder.mlpackage Original encoder, CPU prediction, FP32 exact
PGTDecoder.mlpackage Original temporal decoder, GPU prediction, FP16
manifest.json Algorithm, upstream revisions, SHA-256 of every file
validation.json Split parity against the original network's output tensors
LICENSE, NOTICE Upstream non-commercial licence and attribution

The worker compiles the packages once per macOS release and caches the compiled form, so the first run of a new macOS version pays the compilation cost.

Provenance

Every neural operator is the original network's. The encoder runs on the CPU because its code prediction is a small dense stage that the GPU handles poorly; the decoder runs on the GPU because it is the large temporal stage. This split is what makes the model practical on Apple Silicon without changing its output โ€” validation.json records the measured difference from the original network.

Original network kepengxu/PGTFormer (revision e78a26c) with weights from kepeng/pgtformer-base (revision bc0b7e3).

Licence

Non-commercial. Upstream is BSD-style with a non-commercial restriction. Redistribution is permitted for non-commercial use; commercial use requires permission from Xidian University or the contributors. See LICENSE and NOTICE.

Downloads last month
8
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support