Checking generation

app.py

@@ -86,15 +86,15 @@ def infer(message, history):
 
         combined_embeds = torch.cat(combined_embeds,dim=1)
 
-        #val_combined_embeds = torch.cat([val_image_embeds, img_token_embeds, val_q_embeds], dim=1) # 4, 69, 2560
         predicted_caption = torch.full((1,max_generate_length),50256).to(device)
 
         for g in range(max_generate_length):
-            phi_output_logits = model(inputs_embeds=combined_embeds)['logits'] # 4, 69, 51200
-            predicted_word_token_logits = phi_output_logits[:, -1, :].unsqueeze(1) # 4,1,51200
-            predicted_word_token = torch.argmax(predicted_word_token_logits, dim = -1) # 4,1
+            print(g)
+            phi_output_logits = model(inputs_embeds=combined_embeds)['logits'] 
+            predicted_word_token_logits = phi_output_logits[:, -1, :].unsqueeze(1) 
+            predicted_word_token = torch.argmax(predicted_word_token_logits, dim = -1) 
             predicted_caption[:,g] = predicted_word_token.view(1,-1)
-            next_token_embeds = model.get_input_embeddings()(prompt_tokens) # 4,1,2560
+            next_token_embeds = model.get_input_embeddings()(prompt_tokens) 
             combined_embeds = torch.cat([combined_embeds, next_token_embeds], dim=1)
             
         predicted_captions_decoded = tokenizer.batch_decode(predicted_caption,ignore_index = 50256)[0]