How does it compare to a Markov chain generator I wonder.
replyThe Transformer is the more powerful model than Markov chain, but on such a weak machine as the C64, a MC could output text faster - but it surely would sound "psychedelic", as the memory limits a MC to a first-order or second-order model, so to predict one word, only the two words before would be taken into account as context (and no attention).
replyOn a plain vanilla C64, the Transformer cannot really show what it's capable of doing. An implementation using 2 bit per weight (vectorized) could be slightly better, perhaps.
You can build an unlimited-order Markov chain by, instead of pre-computing a table of counts for all possible contexts, using a substring-search index on the training data to count possible continuations on the fly: https://arxiv.org/abs/2401.17377 That paper uses suffix arrays, but more compact indices are possible: https://arxiv.org/abs/2506.12229
reply[dead]
reply