Update app.py

testing just 800 chars

app.py

@@ -89,7 +89,7 @@ class ModelWrapper:
               prompt = self.tokenizer.apply_chat_template(message, tokenize=False, continue_final_message=True, enable_thinking=False)
         return prompt
 
-    def get_response(self, input, temperature=0.7, top_k=20, top_p=0.8, max_new_tokens=256, enable_thinking=True, system_prompt=SYSTEM_PROMPT):
+    def get_response(self, input, temperature=0.7, top_k=20, top_p=0.8, max_new_tokens=128, enable_thinking=True, system_prompt=SYSTEM_PROMPT):
         """Generate and decode the response with the recommended temperature settings for thinking and non-thinking."""
         print("Generating response...")
         
@@ -136,50 +136,55 @@ model = ModelWrapper(MODEL_NAME)
 # — Gradio inference function —
 def compliance_check(rules_text, transcript_text, thinking):
     try:
+        # Extremely conservative limits to prevent HTTP errors
+        if len(rules_text) > 500:
+            return "Error: Rules text too long. Please limit to 500 characters."
+        
+        if len(transcript_text) > 1000:
+            return "Error: Transcript too long. Please limit to 1000 characters."
+        
         rules = [r for r in rules_text.split("\n") if r.strip()]
         inp = format_rules(rules) + format_transcript(transcript_text)
         
-        # Limit max tokens to prevent oversized responses
-        out = model.get_response(inp, enable_thinking=thinking, max_new_tokens=256)
+        # Very small token limit
+        out = model.get_response(inp, enable_thinking=thinking, max_new_tokens=128)
         
-        # Ensure response isn't too long for HTTP response
-        max_chars = 1500  # Conservative limit
+        # Aggressive truncation - keep response very short
+        max_chars = 800  # Very conservative
         if len(out) > max_chars:
-            out = out[:max_chars] + "\n\n[Response truncated to prevent server errors]"
+            out = out[:max_chars] + "..."
+        
+        # Ensure it's valid UTF-8 and clean
+        out = out.encode('utf-8', errors='ignore').decode('utf-8').strip()
         
-        # Clean up any malformed output
-        out = out.strip()
         if not out:
-            out = "No response generated. Please try with different input."
+            out = "No response generated."
             
         return out
         
     except Exception as e:
-        error_msg = f"Error: {str(e)[:200]}"  # Limit error message length too
-        print(f"Full error: {e}")
-        return error_msg
+        # Very short error message
+        return f"Error: Processing failed"
 
 # — build Gradio interface —
 demo = gr.Interface(
     fn=compliance_check,
     inputs=[
-        gr.Textbox(lines=5, label="Rules (one per line)", max_lines=10),
-        gr.Textbox(lines=10, label="Transcript", max_lines=15),
-        gr.Checkbox(label="Enable ⟨think⟩ mode", value=True)
+        gr.Textbox(lines=3, label="Rules (one per line)", max_lines=5, placeholder="Enter 1-3 rules, keep short"),
+        gr.Textbox(lines=5, label="Transcript", max_lines=8, placeholder="Enter transcript, max 1000 chars"),
+        gr.Checkbox(label="Enable ⟨think⟩ mode", value=False)  # Default to False for faster processing
     ],
-    outputs=gr.Textbox(label="Compliance Output", lines=10, max_lines=15),
+    outputs=gr.Textbox(label="Compliance Output", lines=8, max_lines=10),
     title="DynaGuard Compliance Checker",
-    description="Paste your rules & transcript, then hit Submit.",
-    allow_flagging="never",
+    description="Short inputs only. Rules: max 500 chars. Transcript: max 1000 chars.",
+    flagging_mode="never",
     show_progress=True
 )
 
 if __name__ == "__main__":
-    # Launch with specific server settings to handle response size issues
+    # Minimal launch configuration
     demo.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        show_error=True,
-        quiet=False,
-        max_threads=1  # Limit concurrent requests
+        show_error=False,  # Hide detailed errors
+        quiet=True,        # Reduce logging
+        share=False
     )