Update app.py

app.py

@@ -1,5 +1,5 @@
 """
-Estella Explainer — Hugging Face Space
+Estella Explainer — Hugging Face ZeroGPU Space
 A math reading coach for grades 3–8.
 """
 
@@ -53,21 +53,22 @@ print(f"Loading base model: {BASE_MODEL_ID}")
 
 base_model = AutoModelForCausalLM.from_pretrained(
     BASE_MODEL_ID,
-    torch_dtype=torch.float16,
+    dtype=torch.float16,
     low_cpu_mem_usage=True,
     **HUB_ARGS,
 )
 
-base_model = base_model.to("cuda")
-
-print(f"Loading Estella adapter: {ADAPTER_ID}")
+print(f"Loading adapter on CPU: {ADAPTER_ID}")
 
 model = PeftModel.from_pretrained(
     base_model,
     ADAPTER_ID,
+    torch_device="cpu",
     **HUB_ARGS,
 )
 
+print("Moving combined model to CUDA.")
+
 model = model.to("cuda")
 model.eval()
 
@@ -136,13 +137,11 @@ def ask_estella(message: str, history: list) -> str:
 
     generated_tokens = outputs[0][input_length:]
 
-    hint = tokenizer.decode(
+    return tokenizer.decode(
         generated_tokens,
         skip_special_tokens=True,
     ).strip()
 
-    return hint
-
 
 demo = gr.ChatInterface(
     fn=ask_estella,