LeeSinLiang

def top_k_top_p_filter(logits, top_k: int = 0, top_p: float = 0.0):
    if top_k > 0:
        filter = torch.topk(logits, min(top_k, logits.size(-1)))[0]
        logits[logits < filter[:, [-1]]] = float('-inf')
    if top_p > 0.0:
        sorted_logits, sorted_indices = torch.sort(logits, descending=True)
        cumulative_probs = torch.cumsum(
            F.softmax(sorted_logits, dim=-1), dim=-1)
        filter = cumulative_probs > top_p
        filter[..., 1:] = filter[..., :-1].clone()

LeeSinLiang

import torch.nn as nn

class MultiInputSequential(nn.Sequential):
	def forward(self, *inputs):
		for module in self._modules.values():
			if type(inputs) == tuple:
				inputs = module(*inputs)
			else:
				inputs = module(inputs)
		return inputs