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.
Mixtral 8x7b is out. Is there any plan for RWKV to support MoE in the future, with inference speedup? Looking forward to it.