Update app.py

app.py

@@ -1,3 +1,4 @@
+import os
 import torch
 import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -8,7 +9,9 @@ from peft import PeftModel
 # ---------------------------------
 BASE_MODEL = "meta-llama/Meta-Llama-3-8B-Instruct"
 LORA_PATH = "vastu_lora_adapter_975"
-DEVICE = "cpu"  # FORCE CPU for HF Spaces
+DEVICE = "cpu"
+
+HF_TOKEN = os.getenv("HF_TOKEN")  # 🔥 REQUIRED
 
 SYSTEM_PROMPT = """You are a strict and authoritative Vastu Shastra expert.
 You clearly classify every placement as IDEAL, ACCEPTABLE, or INADVISABLE.
@@ -18,24 +21,32 @@ Avoid unnecessary philosophy. Be precise and actionable.
 """
 
 # ---------------------------------
-# LOAD MODEL (CPU)
+# LOAD MODEL (CPU SAFE)
 # ---------------------------------
 @torch.inference_mode()
 def load_model():
+    tokenizer = AutoTokenizer.from_pretrained(
+        BASE_MODEL,
+        token=HF_TOKEN,
+        trust_remote_code=True,
+    )
+    tokenizer.pad_token = tokenizer.eos_token
+
     base_model = AutoModelForCausalLM.from_pretrained(
         BASE_MODEL,
+        token=HF_TOKEN,                # 🔥 REQUIRED
         torch_dtype=torch.float32,
-        device_map={"": DEVICE},
+        device_map="cpu",              # 🔥 FIXED
         low_cpu_mem_usage=True,
         trust_remote_code=True,
     )
 
-    tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
-    tokenizer.pad_token = tokenizer.eos_token
+    model = PeftModel.from_pretrained(
+        base_model,
+        LORA_PATH,
+    )
 
-    model = PeftModel.from_pretrained(base_model, LORA_PATH)
     model.eval()
-
     return model, tokenizer
 
 
@@ -47,10 +58,8 @@ model, tokenizer = load_model()
 def generate_response(user_prompt):
     prompt = f"""### System:
 {SYSTEM_PROMPT}
-
 ### User:
 {user_prompt}
-
 ### Response:
 """
 
@@ -58,9 +67,9 @@ def generate_response(user_prompt):
 
     outputs = model.generate(
         **inputs,
-        max_new_tokens=128,     # 🔥 reduced
-        do_sample=False,        # 🔥 deterministic
-        temperature=0.3,        # 🔥 stable
+        max_new_tokens=128,
+        do_sample=False,
+        temperature=0.3,
         repetition_penalty=1.1,
         pad_token_id=tokenizer.eos_token_id,
     )
@@ -68,8 +77,9 @@ def generate_response(user_prompt):
     decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
     return decoded.split("### Response:")[-1].strip()
 
+
 # ---------------------------------
-# GRADIO UI (ASYNC CHAT)
+# GRADIO UI
 # ---------------------------------
 with gr.Blocks(title="🧭 Vastu AI Advisor (CPU)") as demo:
     gr.Markdown("# 🧭 Vastu AI Advisor")
@@ -91,4 +101,9 @@ with gr.Blocks(title="🧭 Vastu AI Advisor (CPU)") as demo:
 
     msg.submit(chat, [msg, chatbot], [chatbot, msg])
 
-demo.launch()
+
+demo.launch(
+    server_name="0.0.0.0",   # 🔥 REQUIRED FOR HF SPACES
+    server_port=7860,
+)
+