RWKV Weights

Current release: G1k for RWKV Chat 4.8.5

The current release contains 40 published artifacts from the 2026-09-30 G1k checkpoints with context length 25,600. Both repositories contain the same file sizes and SHA-256 identities.

MLX

Model	Quantization	Supported platforms	Download
1.5B	W6 / group 64	macOS, iOS	ModelScope · Hugging Face
2.9B	W6 / group 64	macOS, iOS	ModelScope · Hugging Face
7.2B	W6 / group 64	macOS, iOS	ModelScope · Hugging Face
13.3B	W6 / group 64	macOS	ModelScope · Hugging Face

These MLX packages include the configuration used by the Swift runtime. Select a package matching your platform and available memory.

Complete artifact catalog

The G1k release manifest (ModelScope · Hugging Face) lists all 40 files, quantization recipes, platform and chip limits, sizes, hashes, and frozen mirror revisions. It covers MLX, CoreML, llama.cpp, WebRWKV, Palm, QNN and MediaTek artifacts according to the existing supported combinations.

Use index.json for discovery (ModelScope · Hugging Face). Its latestManifest points to the current G1k release. The App consumer catalog is recorded in the release manifest.

Historical releases

Older immutable artifacts and manifests are retained for reproducibility. The previous G1i download notes are archived (ModelScope · Hugging Face). Superseded G1i/G1j MLX entries are removed from the current 4.8.5 App catalog.

Repository layout

artifacts/<family>/<task>/<size>/<backend>/<artifact>
manifests/<family>/<task>/<immutable-manifest>.json
index.json

Downloads last month: 806

GGUF

Model size

2B params

Architecture

rwkv7

Hardware compatibility

4-bit

6-bit

8-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support