update app

app.py

@@ -6,6 +6,20 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 # Model configuration
 MODEL_NAME = "Text2MotionPrompter/Text2MotionPrompter"
 
+# Pre-load tokenizer at startup
+print("📥 Loading tokenizer...")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+print("✅ Tokenizer loaded successfully!")
+
+# Pre-load model to CPU at startup
+print(f"📥 Loading model to CPU: {MODEL_NAME}...")
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME,
+    torch_dtype="auto",
+    device_map="cpu"
+)
+print("✅ Model loaded to CPU successfully!")
+
 # Prompt template
 TEMPLATE = """
 # Role
@@ -46,25 +60,6 @@ Analyze the user-provided [Input Action] and generate a structured JSON response
 {}
 """
 
-# Global variables for lazy loading
-tokenizer = None
-model = None
-
-
-def load_model():
-    """Load the model and tokenizer (lazy loading)"""
-    global tokenizer, model
-    if tokenizer is None:
-        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-    if model is None:
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_NAME,
-            torch_dtype=torch.float16,
-            device_map="auto"
-        )
-    return tokenizer, model
-
-
 @spaces.GPU(duration=120)
 def generate_motion_prompt(action_input: str, max_new_tokens: int = 512) -> str:
     """
@@ -80,8 +75,8 @@ def generate_motion_prompt(action_input: str, max_new_tokens: int = 512) -> str:
     if not action_input.strip():
         return "Please enter an action description."
     
-    # Load model (will use cached version if already loaded)
-    tokenizer, model = load_model()
+    # Move model to GPU
+    model.to("cuda")
     
     # Prepare the prompt
     messages = [
@@ -94,7 +89,7 @@ def generate_motion_prompt(action_input: str, max_new_tokens: int = 512) -> str:
         add_generation_prompt=True,
     )
     
-    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+    model_inputs = tokenizer([text], return_tensors="pt").to("cuda")
     
     # Generate response
     with torch.no_grad():