Pre-baked CoreML models: skip the coremltools headache

What it does This repository acts as a phone book for pre-converted CoreML models. It catalogs dozens of popular architectures—YOLO variants, Stable Diffusion checkpoints, vision transformers, audio separators, and even small LLMs like Gemma and Qwen—each with a Google Drive link to the converted weights and the original license. A few entries point to companion sample projects showing how to wire the model into an iOS app.

The interesting bit The real product here is grunt work. Turning PyTorch or TensorFlow weights into CoreML-compatible graphs is tedious and error-prone; this zoo collects the results so Apple-ecosystem developers can skip straight to bundling. The breadth is almost comical: object detectors, anime-style transfer networks, speaker diarization, and text-to-music generators share the same page.

Key highlights

• Covers vision, audio, and language: from EfficientNet to Florence-2, HTDemucs, and Kokoro TTS.
• Includes recent heavy hitters like YOLO11, RF-DETR, Depth Anything 3, and several Stable Diffusion v1.5 fine-tunes.
• Most downloads are direct Google Drive links; some newer entries include sample Xcode projects (e.g., D-FINE and RF-DETR demos).
• Each model retains its original license and attribution link.

Caveats

• This is a catalog, not a framework: there is no unified API, conversion script, or versioning system for the zoo itself.
• Distribution relies on external Google Drive links, so longevity and reproducibility depend on the maintainer’s storage.
• README notes for some models (e.g., D-FINE) warn that post-processing like NMS is left as an exercise for the developer.

Verdict Worth bookmarking if you build iOS or macOS features with on-device ML and want to prototype fast without wrestling with coremltools. Skip it if you need training pipelines, guaranteed long-term artifact hosting, or a programmatic package manager.