Update app.py

app.py

@@ -233,20 +233,24 @@ Focus on clear, concise, and evidence-based improvements that align with the ove
             add_generation_prompt=True
         )
         
-        # Check input length and truncate to first 10k tokens
+        # Check input length and truncate to 15000 tokens before encoding
         input_tokens = tokenizer.encode(text)
-        if len(input_tokens) > 10000:  # Limit to 10k tokens as requested
-            input_tokens = input_tokens[:10000]
+        if len(input_tokens) > 15000:  # Limit to 15k tokens
+            input_tokens = input_tokens[:15000]
             text = tokenizer.decode(input_tokens)
-            print(f"Input truncated to 10000 tokens")
+            print(f"Input truncated to 15000 tokens")
         
         progress(0.5, desc="Generating improved text...")
         # Generate non-streaming
         input_ids = tokenizer.encode(text, return_tensors="pt").to(model.device)
         
+        # Create attention mask
+        attention_mask = torch.ones_like(input_ids)
+        
         with torch.no_grad():
             output_ids = model.generate(
                 input_ids,
+                attention_mask=attention_mask,  # 添加注意力掩码
                 max_new_tokens=max_new_tokens,
                 do_sample=(temperature > 0),
                 temperature=temperature if temperature > 0 else 1.0,