RWKV mobile model pack (WebRWKV + GGUF)
This repo is less “one model” and more a distribution hub for RWKV-family weights in formats that are convenient on constrained devices. It includes WebRWKV artifacts (you’ll see lots of .st and .prefab files under WebRWKV/) across multiple sizes and versions, and also GGUF files (e.g. under gguf/) for people who want to run RWKV variants via llama.cpp-style tooling.
The interesting part is the emphasis on practical formats and quantizations: you can choose smaller models for on-device experiments, or grab NF4 / int8-style variants where the runtime supports them. There are also some “utility” subfolders (including ONNX artifacts under multimodal/) that hint at broader mobile inference workflows beyond plain text.
If you want a quick first try, pick a small RWKV v7 weight (for example a 0.4B or 1.5B file with a context length you can afford) and load it in a compatible runner: WebRWKV for the .st/.prefab builds, or a GGUF-capable runner for the GGUF variants. Expect some trial-and-error: format support and context length are both runner-specific, so match the file type to the app/library you’re actually using.
Quick stats from the listing feed: 12 likes · 7807 downloads.
Source listing: https://huggingface.co/models?sort=modified