help="Path of the corpus to evaluate on", default=None) argp.add_argument('--outputs_path', default=None) args = argp.parse_args() # Save the device device = torch.cuda.current_device() if torch.cuda.is_available() else 'cpu' # Keep the block size 128 # Why is the pretraining corpus always required (even if we're not pretraining?) # It's because we're using it as a hack to always have the same vocabulary # (that is, the same mapping from character to integer, and we build the # vocab from the pretraining corpus.) block_size = 128 text = open(args.pretrain_corpus_path).read() pretrain_dataset = dataset.CharCorruptionDataset(text, block_size) # We don't suggest you change these hyperparameters, as they're known to work. # use them for both the vanilla and the synthesizer models mconf = model.GPTConfig(pretrain_dataset.vocab_size, pretrain_dataset.block_size, n_layer=4, n_head=8, n_embd=256) """ Don't change above here; write your code below """ if args.variant == 'vanilla': pass # TODO [part c]: Make some model here elif args.variant == 'synthesizer':
argp.add_argument('--eval_corpus_path', help="Path of the corpus to evaluate on", default=None) argp.add_argument('--outputs_path', default=None) args = argp.parse_args() # Save the device device = torch.cuda.current_device() if torch.cuda.is_available() else 'cpu' # Keep the block size 128 # Why is the pretraining corpus always required (even if we're not pretraining?) # It's because we're using it as a hack to always have the same vocabulary # (that is, the same mapping from character to integer, and we build the # vocab from the pretraining corpus.) block_size = 128 text = open(args.pretrain_corpus_path).read() pretrain_dataset = dataset.CharCorruptionDataset(text, block_size) # We don't suggest you change these hyperparameters, as they're known to work. # use them for both the vanilla and the synthesizer models mconf = model.GPTConfig(pretrain_dataset.vocab_size, pretrain_dataset.block_size, n_layer=4, n_head=8, n_embd=256) """ Don't change above here; write your code below """ if args.variant == 'vanilla': #pass # TODO [part c]: Make some model here model = model.GPT(mconf, "vanilla") elif args.variant == 'synthesizer': #pass # TODO [part g]: Make some other model here