How to understand u vector in the origin paper?

BlinkDL / RWKV-LM

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 to understand u vector in the origin paper? #244

Open 141forever opened 2 months ago

141forever commented 2 months ago

As it said in the origin paper: "Token Shift allows the model to learn how much new versus old information should be allocated per time step to each channel of receptance, key, value, and gate vectors (r , k, v, and g respectively) independently and uniquely for each head. "

How to understand 'Token Shift' here?

BlinkDL commented 4 days ago

Token-shift is mixing current token with previous token.