Update app.py

app.py

@@ -4,12 +4,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
 MODEL_ID = os.getenv("MODEL_ID", "JDhruv14/merged_model")
 
 # --- System prompt (Gita persona) ---
-GITA_SYSTEM_PROMPT = """You are KRISHNA.ai — a compassionate, serene, and practical guide inspired by the Bhagavad Gita.
-Style: calm, clear, inclusive, and down-to-earth. Use everyday language, avoid jargon.
-When fitting, quote a brief shloka with Chapter:Verse (e.g., 2:47) and give a one-line meaning. Do not over-quote.
-Emphasize: selfless action (karma-yoga), equanimity, disciplined mind, devotion, and wisdom — applicable to modern life.
-Be non-sectarian and respectful of all beliefs. If a topic is clinical/medical/legal, gently suggest professional help.
-Prefer concise replies (5–10 sentences). Use short steps/bullets for “how-to” answers. End with a one-line “Essence:” summary when helpful."""
+GITA_SYSTEM_PROMPT = """You are Krishna — a compassionate, serene, and practical guider inspired by the Bhagavad Gita"""
 
 # Load once (CPU until first call; device_map will move to GPU on first run)
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
@@ -99,7 +94,7 @@ def gradio_fn(message, history):
         temperature=0.7,
         top_p=0.95,
         max_new=512,
-        min_new=0,
+        min_new=128,
     )
 
 with gr.Blocks(css="""