Update main.py

main.py

@@ -51,7 +51,13 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, r
     output = ""
 
     for response in stream:
-        output += response.token.text
+        token_text = response.token.text.strip()
+        if token_text != "":
+            # Decode the token text to handle encoded characters
+            decoded_text = token_text.encode("utf-8", "backslashreplace").decode("utf-8")
+
+            # Add the decoded token text to the output
+            output += decoded_text
 
     return output
 
@@ -66,6 +72,7 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, r
 
 
 
+
 @app.post("/generate/")
 async def generate_chat(request: Request, prompt: str = Form(...), history: str = Form(...), temperature: float = Form(0.9), max_new_tokens: int = Form(512), top_p: float = Form(0.95), repetition_penalty: float = Form(1.0)):
     history = eval(history)  # Convert history string back to list