Update app.py

app.py

@@ -35,9 +35,8 @@ def load_image_model():
 
 # Load text model on CPU
 def load_text_model():
-    base_model = AutoModelForCausalLM.from_pretrained("google/gemma-2-2b")
-    model = PeftModel.from_pretrained(base_model, "soonbob/gemma-2-2b-tarot")
-    tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-2b")
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
+    model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
     return model, tokenizer
 
 # Generate card description with ZeroGPU
@@ -92,11 +91,24 @@ Provide a professional interpretation covering:
 - Combined message and symbolism
 - Practical advice
 - Potential outcomes"""
-    input_ids = tokenizer(prompt, return_tensors="pt").to("cuda")
-    outputs = model.generate(**input_ids, max_new_tokens=500)
+
+    messages = [
+        {"role": "system", "content": "You are a Tarot Card Explainer provideing relevant suggestions based on tarot card name"},
+        {"role": "user", "content": prompt}
+    ]
+    
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    model_inputs = tokenizer([text], return_tensors="pt").to("cuda")
+    generated_ids = model.generate(**model_inputs, max_new_tokens=512)
+    generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
+    
     # Move output to CPU before decoding
-    interpretation = tokenizer.decode(outputs[0].cpu(), skip_special_tokens=True)
-    return interpretation
+    response = tokenizer.batch_decode(generated_ids.cpu(), skip_special_tokens=True)[0]
+    return response
 
 def main():
     """