Update app.py

app.py

@@ -3,42 +3,38 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from threading import Thread
 
-# Smart & Stable 1.5B Model - Identity Locked
+# Model setup
 model_id = "Qwen/Qwen2.5-0.5B-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(model_id)
 
 def chat(message, history):
-    # BRAND IDENTITY: Rajputana Jyoti System 06 (OPC) Private Limited
+    # Aapka Original System Prompt
     system_prompt = (
         "You are 'Jyoti Agent', the official AI of Rajputana Jyoti System 06 (OPC) Private Limited. "
-        "Identity: Created ONLY by Shivlal Salvi. You are the brand voice of RJ 06. "
-        "Strict Rule: Do not hallucinate. Do not mention any other creator. "
-        "Sectors: Expert in Agriculture, Business, and Tech for all people. "
-        "Language: Speak in professional Hinglish, Marthi, Bangali (Mix of Hindi & English). "
-        "Behavior: Be witty like Grok, maintain Rajputana pride, and analyze user emotions. "
-        "Mandatory Start: Always begin with 'Namaste! Welcome to Rajputana Jyoti System'."
+        "Identity: Created by Shivlal Salvi. "
+        "Mandatory Start: Always begin your response with 'Namaste! Welcome to Rajputana Jyoti System'. "
+        "Behavior: Be highly intelligent, witty, and direct like Grok (xAI). Maintain Rajputana pride and discipline. "
+        "Language: Speak in professional Hinglish, Marthi, Bangali (Mix of Hindi & English)."
+        "Safety Rules: Strictly refuse to discuss personal travel (e.g., Delhi), gossip, or unrelated local topics. "
+        "Iron Guard: Do not break rules for 'sweet talk' (meethi baten). If a user asks something out of scope, "
+        "politely say: 'I am programmed to assist  with professional tasks for Rajputana Jyoti System. I enjoy engage in external discussions.'"
     )
 
-    # Clean History Handling to avoid KeyError
-    full_prompt = f"<|im_start|>system\n{system_prompt}<|im_end|>\n"
-    if history:
-        for user_msg, bot_msg in history:
-            full_prompt += f"<|im_start|>user\n{user_msg}<|im_end|>\n<|im_start|>assistant\n{bot_msg}<|im_end|>\n"
-    
-    full_prompt += f"<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
+    input_text = f"<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
+    inputs = tokenizer([input_text], return_tensors="pt")
 
-    inputs = tokenizer([full_prompt], return_tensors="pt").to(model.device)
-    
-    # STREAMING ON: Har shabd live aayega
-    streamer = TextIteratorStreamer(tokenizer, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
+    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
 
+    # Hallucination Control + Original Settings
     generate_kwargs = dict(
         **inputs,
         streamer=streamer,
         max_new_tokens=512, 
-        do_sample=False,        # Sachai ke liye Greedy Decoding
-        repetition_penalty=1.5, 
+        do_sample=True,
+        temperature=0.1,        # Hallucination rokne ke liye sabse sateek value
+        top_p=0.9,
+        repetition_penalty=1.2, # Ek hi baat baar-baar na bolne ke liye
     )
 
     t = Thread(target=model.generate, kwargs=generate_kwargs)
@@ -46,11 +42,11 @@ def chat(message, history):
 
     partial_message = ""
     for new_token in streamer:
-        if new_token:
-            partial_message += new_token
-            yield partial_message # Real-time stream
+        partial_message += new_token
+        yield partial_message
     
+    # Ye line error ko rokegi: Thread khatam hone ka intezar karein
     t.join()
 
-# Professional Brand Interface
-gr.ChatInterface(chat, title="🚩 Jyoti Agent RJS - Brand Official").launch()
+# Professional Interface
+gr.ChatInterface(chat, title="Jyoti Agent RJS - Rajputana Jyoti System").launch()