Update app.py

app.py

@@ -1,23 +1,42 @@
-from transformers import AutoModelForCausalLM, AutoTokenizer
+import os
 import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
+from threading import Thread
+import gradio as gr
 
-model_id = "google/gemma-1.1-7b-it"
+# Configuration
+MODEL_ID = "google/gemma-1.1-7b-it"
+HF_TOKEN = os.getenv("HF_TOKEN")  # Will be injected from Space secrets
+MAX_TOKENS = 300
+TEMPERATURE = 0.7
 
-# CPU-specific config
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    device_map="cpu",
-    torch_dtype=torch.float32,  # Required for CPU
-    load_in_8bit=True  # Reduces RAM usage by 2x
-)
+# Authentication check
+if not HF_TOKEN:
+    raise ValueError("""
+    ❌ HF_TOKEN not found! 
+    Add it in Space Settings -> Repository secrets:
+    1. Click Settings ⚙️
+    2. Go to 'Variables and secrets'
+    3. Add new secret: Name=HF_TOKEN, Value=your_hf_token_here
+    """)
 
-def generate(prompt):
-    inputs = tokenizer(prompt, return_tensors="pt").to("cpu")
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=150,  # Must stay under 200
-        do_sample=True,
-        temperature=0.7
+# Load model 
+def load_model():
+    print("🚀 Loading model...")
+    tokenizer = AutoTokenizer.from_pretrained(
+        MODEL_ID,
+        token=HF_TOKEN
     )
-    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        device_map="auto",
+        torch_dtype=torch.float16,
+        token=HF_TOKEN
+    )
+    print("✅ Model loaded!")
+    return tokenizer, model
+
+tokenizer, model = load_model()
+
+# ... [rest of your existing code remains exactly the same] ...