Update app.py

app.py

@@ -2,55 +2,55 @@ import gradio as gr
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from threading import Thread
-import time
 
-# Model setup - 3B is heavy, giving it more breathing room
-model_id = "Qwen/Qwen2.5-3B-Instruct"
+# 1.5B Model: Smart enough to understand, light enough to run stable
+model_id = "Qwen/Qwen2.5-1.5B-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
 
 def chat(message, history):
-    # Strict Identity & Multi-Sector Locked Prompt
+    # IDENTITY & MULTI-SECTOR MASTER PROMPT
     system_prompt = (
         "Your name is 'Jyoti Agent', official AI of Rajputana Jyoti System 06 (OPC) Private Limited. "
-        "Founder: Shivlal Salvi. Identity Lock: Created ONLY by Rajputana Jyoti System 06. "
-        "Behavior: Be witty like Elon Musk's Grok. Analyze user emotions (Sad/Happy). "
-        "Expertise: Master in Agriculture, Business, and Tech. "
+        "Identity: Created ONLY by Shivlal Salvi. No other creators exist. "
+        "Strict Rule: Do not hallucinate. Do not mention 'Yashraj' or 'Dr. Ajit Jain'. "
+        "Sectors: You are an expert in Agriculture, Business, Tech, and Rural development. "
+        "Behavior: Be witty like Grok, maintain Rajputana pride, and analyze user emotions. "
         "Language: Speak Hinglish, Hindi, Marathi, and Bengali. "
-        "Start: Always start with 'Namaste! Welcome to Rajputana Jyoti System'."
+        "Mandatory Start: Always begin with 'Namaste! Welcome to Rajputana Jyoti System'."
     )
 
+    # Context management for memory
     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"
 
-    inputs = tokenizer([full_prompt], return_tensors="pt")
+    inputs = tokenizer([full_prompt], return_tensors="pt").to(model.device)
     
-    # Timeout ko 20.0 se badha kar 60.0 kiya hai taaki bade model ko time mile
-    streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
+    # Fast Streamer
+    streamer = TextIteratorStreamer(tokenizer, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
 
+    # ANTI-HALLUCINATION SETTINGS
     generate_kwargs = dict(
         **inputs,
         streamer=streamer,
         max_new_tokens=512, 
-        do_sample=False,
-        repetition_penalty=1.4, 
+        do_sample=False,        # Greedy decoding: No 'random' words allowed
+        repetition_penalty=1.5, # Stops the AI from getting stuck in loops
     )
 
     t = Thread(target=model.generate, kwargs=generate_kwargs)
     t.start()
 
+    # Streaming the output
     partial_message = ""
-    try:
-        for new_token in streamer:
-            if new_token:
-                partial_message += new_token
-                yield partial_message
-    except Exception as e:
-        yield partial_message + "\n\n[System Note: Stream timed out, but Identity is safe.]"
+    for new_token in streamer:
+        if new_token:
+            partial_message += new_token
+            yield partial_message
     
     t.join()
 
 # Professional Interface
-gr.ChatInterface(chat, title="🚩 Jyoti Agent RJS - 3B Stable Unit").launch()
+gr.ChatInterface(chat, title="🚩 Jyoti Agent RJS - Balanced Intelligence").launch()