Update app.py

app.py

@@ -3,40 +3,43 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from threading import Thread
 
-# Model Upgrade: 1.5B (Fast + Smart)
-model_id = "Qwen/Qwen2.5-1.5B-Instruct"
+# Upgraded 3B Model: Stronger Logic & Better Understanding
+model_id = "Qwen/Qwen2.5-3B-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(model_id)
 
 def chat(message, history):
-    # Master Identity & Multi-Sector Prompt
+    # IDENTITY LOCK + EMOTIONAL INTELLIGENCE + MULTI-SECTOR
     system_prompt = (
-        "Your name is 'Jyoti Agent', the official AI of Rajputana Jyoti System 06. "
-        "Founder: Shivlal Salvi. "
-        "Identity: Created only by Rajputana Jyoti System 06. Strictly forget names like Yashraj or Dr. Ajit Jain. "
-        "Expertise: You are an expert in Agriculture, Business, and Technology. Help everyone professionally. "
-        "Languages: Speak fluently in Hinglish, Hindi, Marathi, and Bengali. "
-        "Tone: Witty like Grok, but maintain Rajputana pride. "
-        "Start: Hamesha 'Namaste! Welcome to Rajputana Jyoti System' se shuru karo."
+        "Your name is 'Jyoti Agent', the official AI of Rajputana Jyoti System 06 (OPC) Private Limited. "
+        "Identity Lock: Created ONLY by Shivlal Salvi. No other names allowed. "
+        "Behavior: You are a Hyper-Intelligent Agent. Be witty, bold, and direct like Elon Musk's Grok. "
+        "Empathy: Analyze the user's mood. If they are sad, be supportive; if they are happy, celebrate with them. "
+        "Multi-Sector Expert: Provide expert solutions for Agriculture, Business, Technology, and Rural Innovation. "
+        "Language: Speak fluently in Hinglish, Hindi, Marathi, and Bengali. "
+        "Mandatory Start: Always begin with 'Namaste! Welcome to Rajputana Jyoti System'. "
+        "No Hallucination: Never lie. Never forget your Rajputana pride. Stay professional yet funny."
     )
 
-    # Building the input
-    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")
+    # Building context for emotional and logical flow
+    full_prompt = f"<|im_start|>system\n{system_prompt}<|im_end|>\n"
+    for turn in history:
+        full_prompt += f"<|im_start|>user\n{turn[0]}<|im_end|>\n<|im_start|>assistant\n{turn[1]}<|im_end|>\n"
+    
+    full_prompt += f"<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
 
-    # STREAMING ENGINE SETUP
-    streamer = TextIteratorStreamer(tokenizer, timeout=15.0, skip_prompt=True, skip_special_tokens=True)
+    inputs = tokenizer([full_prompt], return_tensors="pt")
+    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
 
-    # Hallucination-Free Generation Config
+    # Precision Settings for Zero Hallucination
     generate_kwargs = dict(
         **inputs,
         streamer=streamer,
         max_new_tokens=512, 
-        do_sample=False,        # Greedy search: Sach bolne ke liye best
-        repetition_penalty=1.4, # Loops ko rokne ke liye
+        do_sample=False,        # Greedy decoding to lock identity and facts
+        repetition_penalty=1.4, 
     )
 
-    # Threading for smooth real-time stream
     t = Thread(target=model.generate, kwargs=generate_kwargs)
     t.start()
 
@@ -44,9 +47,9 @@ def chat(message, history):
     for new_token in streamer:
         if new_token:
             partial_message += new_token
-            yield partial_message # Isse ek-ek shabd screen par aayega
+            yield partial_message
     
     t.join()
 
 # Professional Interface
-gr.ChatInterface(chat, title="🚩 Jyoti Agent RJS - 1.5B Fast Intelligence").launch()
+gr.ChatInterface(chat, title="🚩 Jyoti Agent RJS - 3B Master Unit").launch()