AGENDD / RWKV-ASR

This repo is an exploratory experiment to enable frozen pretrained RWKV language models to accept speech modality input. We followed the idea of SLAM_ASR and used the RWKV language model as the LLM, and instead of directly writing a prompt template we directly finetuned the initial state of the RWKV model.
32Updated last week

Related projects

Alternatives and complementary repositories for RWKV-ASR