Update app.py

app.py

@@ -27,11 +27,10 @@ def generate_text(usertitle, content, temperature, max_length, N=3):
     # 'content': content
     # }
     input_text = f"[[[title:]]] {usertitle}\n[[[content:]]]{content}\n\n"
-    inputs = tokenizer.apply_chat_template(input_text, return_tensors='pt').cuda()
+    inputs = tokenizer(input_text, return_tensors='pt').to('cuda')
     attention_mask = torch.ones(inputs['input_ids'].shape, dtype=torch.long, device='cuda')
     generated_sequences = model.generate(inputs['input_ids'], attention_mask=attention_mask, temperature=temperature, max_length=max_length, pad_token_id=tokenizer.eos_token_id, num_return_sequences=N, do_sample=True)
-    decoded_sequences = [tokenizer.decode(g) for g in generated_sequences]#.strip().split(tokenizer.eos_token)[0]
-
+    decoded_sequences = [tokenizer.decode(g) for g in generated_sequences]
     def score(sequence):
         inputs = rm_tokenizer(sequence, return_tensors='pt', padding=True, truncation=True, max_length=512).to('cuda')
         inputs = {k: v.to('cuda') for k, v in inputs.items()} 
@@ -42,7 +41,7 @@ def generate_text(usertitle, content, temperature, max_length, N=3):
             logits = outputs.logits
             print("Logits shape:", logits.shape)  
             print("Logits contents:", logits) 
-            return logits[0]#[0].item()
+            return logits[0]
 
     best_sequence = max(decoded_sequences, key=score)