where is "Puncond"?

Where is the “Puncond”in the original paper, please!

This author writes this code clearly what I have seen. I think the Puncond is here in training

Line 249 in 3b7aa12

    
           context_mask = torch.bernoulli(torch.zeros_like(c)+self.drop_prob).to(self.device)

Sampling is here

Conditional_Diffusion_MNIST/script.py

Lines 266 to 272 in 3b7aa12

    
           context_mask = torch.zeros_like(c_i).to(device) 
        
           # double the batch 
        
           c_i = c_i.repeat(2) 
        
           context_mask = context_mask.repeat(2) 
        
           context_mask[n_sample:] = 1. # makes second half of batch context free

Indeed, self.drop_prob is the probability of unconditional training.

	context_mask = torch.zeros_like(c_i).to(device)

	# double the batch
	c_i = c_i.repeat(2)
	context_mask = context_mask.repeat(2)
	context_mask[n_sample:] = 1. # makes second half of batch context free