RWKV is an RNN with transformer-level LLM performance. It can be directly trained like a GPT (parallelizable). So it's combining the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding.
Apache License 2.0
11.99k
stars
825
forks
source link
How does the generation speed of RWKV-5/6 compare to that of mamba with the same number of parameters? #236
similar. try https://github.com/BlinkDL/ChatRWKV/blob/main/API_DEMO_CHAT.py