) # Needed to initialize the model's cache
for s in range(to_generate.min(), to_generate.max() + 1):
output = self(BracketedSequence(input, s, 1)).x
- logits = output[:, s]
+ logits = output[:, s] / temperature
if forbidden_tokens is not None:
logits = logits.masked_fill(forbidden_tokens, float("-inf"))
if forced_biases is not None:
t_next = dist.sample()
sum_logits += logits.log_softmax(dim=-1)[
torch.arange(t_next.size(0)), t_next
- ]
+ ].sum()
input[:, s] = ar_mask[:, s] * t_next + (1 - ar_mask[:, s]) * input[:, s]
return sum_logits