English | 中文

Based on rwkv.cpp, supports running RWKV4, RWKV5, RWKV6, RWKV7 World/Raven/Finch 1B5-14B (ggml) (CPU/GPU)

Installation

Install from nuget.org

PM> Install-Package RWKVSharp

If you are running on Windows CPU, you also need to install the following:

PM> Install-Package RWKVSharp.Native.Cpu.avx2.win-x64

For native libraries on other systems, you can search at nuget.org

How to use RWKVSharp

using RWKVSharp.Core;
using RWKVSharp;
var rf = new RunnerFactory();
rf.RegisterRWKVGGMLModel("RWKV-x070-World-0.1B-v2.8-20241210-ctx4096-FP16.bin", "rwkv_vocab_v20230424.txt");
// RWKV-x070-World-0.1B-v2.8-20241210-ctx4096-FP16.bin conversion instructions are detailed below.
// rwkv_vocab_v20230424.txt can be downloaded from https://github.com/imxcstar/RWKVSharp/tree/main/RWKVSharp.Test/Model
var r = rf.Builder();
while (true)
{
    Console.Write(">");
    var value = Console.ReadLine();
    if (string.IsNullOrEmpty(value))
        continue;
    r.Generate(value.Replace("\\r\\n", "\r\n").Replace("\\r", "\r").Replace("\\n", "\n"), Console.Write);
    Console.WriteLine();
}

Model conversion method

1. First, download the .pth model you wish to use from https://huggingface.co/RWKV
2. Then, download the RWKVSharp.Convert.zip conversion tool from Releases (currently only supports Windows)
3. Use the conversion tool to convert the model.
4. After conversion, you'll obtain a .bin model file that can be used with RWKVSharp.

Model Quantization Tool

You can download RWKVSharp.Quantize.zip from Releases (currently only supports Windows).

Preview

Links

• RWKV Hugging Face repository: https://huggingface.co/RWKV
• rwkv.cpp: https://github.com/RWKV/rwkv.cpp
• RWKV: https://github.com/BlinkDL/RWKV-LM
• ChatRWKV: https://github.com/BlinkDL/ChatRWKV
• Tokenizer: https://github.com/Alex1911-Jiang/GPT-3-Encoder-Sharp