Upload app.py with huggingface_hub

app.py

@@ -4,29 +4,26 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 
-# Lazy load model and tokenizer on first request (more reliable on Spaces)
-MODEL = None
-TOKENIZER = None
-
-def get_model():
-    global MODEL, TOKENIZER
-    if MODEL is None or TOKENIZER is None:
-        print("Loading model...")
-        base_model = AutoModelForCausalLM.from_pretrained(
-            "HuggingFaceTB/SmolLM-360M",
-            torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
-            device_map="auto"
-        )
-        tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-360M")
-        if tokenizer.pad_token is None:
-            tokenizer.pad_token = tokenizer.eos_token
+# Load model and tokenizer
+@gr.utils.async_cache
+def load_model():
+    print("Loading model...")
+    base_model = AutoModelForCausalLM.from_pretrained(
+        "HuggingFaceTB/SmolLM-360M",
+        torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+        device_map="auto"
+    )
+    tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-360M")
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    
+    # Load LoRA adapters
+    model = PeftModel.from_pretrained(base_model, "waliaMuskaan011/calendar-event-extractor-smollm")
+    model.eval()
+    print("Model loaded successfully!")
+    return model, tokenizer
 
-        # Load LoRA adapters
-        model = PeftModel.from_pretrained(base_model, "waliaMuskaan011/calendar-event-extractor-smollm")
-        model.eval()
-        MODEL, TOKENIZER = model, tokenizer
-        print("Model loaded successfully!")
-    return MODEL, TOKENIZER
+model, tokenizer = load_model()
 
 def extract_calendar_event(event_text):
     """Extract calendar information from natural language text."""
@@ -34,8 +31,6 @@ def extract_calendar_event(event_text):
     if not event_text.strip():
         return "Please enter some text describing a calendar event."
     
-    model, tokenizer = get_model()
-    
     # Build prompt
     prompt = f"""Extract calendar fields from: "{event_text}".
 Return ONLY valid JSON with keys [action,date,time,attendees,location,duration,recurrence,notes].