Update handler.py

handler.py

@@ -7,7 +7,7 @@ from huggingface_hub import snapshot_download
 BASE_MODEL = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"
 ADAPTER_PATH = "GilbertAkham/deepseek-R1-multitask-lora"
 
-# === System message (always prepended to input) ===
+# === System message ===
 SYSTEM_PROMPT = (
     "You are Chat-Bot, a helpful and logical assistant trained for reasoning, "
     "email, chatting, summarization, story continuation, and report writing.\n\n"
@@ -37,8 +37,8 @@ class EndpointHandler:
 
     def __call__(self, data):
         # === Combine system + user prompt ===
-        prompt = data.get("inputs", "")
-        full_prompt = SYSTEM_PROMPT + prompt
+        user_prompt = data.get("inputs", "")
+        full_prompt = SYSTEM_PROMPT + user_prompt
 
         params = data.get("parameters", {})
         max_new_tokens = params.get("max_new_tokens", 512)
@@ -58,5 +58,9 @@ class EndpointHandler:
                 eos_token_id=self.tokenizer.eos_token_id,
             )
 
+        # === Decode and strip system message ===
         text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+        if text.startswith(SYSTEM_PROMPT):
+            text = text[len(SYSTEM_PROMPT):].strip()
+
         return {"generated_text": text}