ggerganov's picture
talk-llama : use llama_decode instead of llama_eval
301b000 unverified