ChatRWKV
A ChatGPT-like conversational interface powered by RWKV, a fully recurrent (RNN) language model that matches transformer quality while running faster and using less memory.
ChatRWKV is a ChatGPT-like chat interface powered by RWKV, a fully recurrent neural network (RNN) language model. RWKV is presented as an RNN that can match transformer-based models in quality and scaling while running faster and using less VRAM.
The related RWKV-LM repository holds the model's training and fine-tuning code and explanation, and related projects include Albatross for efficient inference, an Android/iOS app for local inference, a Vulkan-based GPU inference server, a CPU/cuBLAS/CLBlast inference engine supporting int4/int8/fp16/fp32, a LoRA/PiSSA/QLoRA fine-tuning toolkit, and an infinite-context trainer. Models are distributed through Hugging Face in several formats, including GGUF and Hugging Face-compatible weights, and a CUDA kernel can be built for faster, more memory-efficient inference.
It is written in Python and released under the Apache-2.0 license.