* reserve correct size for logits * add functions to get and set the whole llama state: including rng, logits, embedding and kv_cache * remove unused variables * remove trailing whitespace * fix comment
