GPT (Generative Pre-trained Transformer)

KV Cache

Inference optimization that caches the keys and values of previous tokens to avoid recalculating attention states at each new token generation.

← Kembali