move to cpu runtime

app.py

@@ -1,4 +1,3 @@
-import spaces
 import gradio as gr
 from model import DecoderTransformer, Tokenizer
 from huggingface_hub import hf_hub_download
@@ -12,7 +11,7 @@ n_layer=6
 n_head=6
 dropout=0.2
 
-device = 'cuda'
+device = 'cpu'
 
 model_id = "philipp-zettl/chessPT"
 
@@ -24,7 +23,6 @@ model.load_state_dict(torch.load(model_path, map_location=torch.device('cpu')))
 model.to(device)
 tokenizer = Tokenizer.from_pretrained(tokenizer_path)
 
-@spaces.GPU
 def greet(prompt):
     model_input = torch.tensor(tokenizer.encode(prompt), dtype=torch.long, device=device).view((1, len(prompt)))
     return tokenizer.decode(model.generate(model_input, max_new_tokens=4, context_size=context_size)[0].tolist())
@@ -32,9 +30,10 @@ def greet(prompt):
 
 with gr.Blocks() as demo:
     gr.Markdown("""
+    # ChessPT
     Welcome to ChessPT.
 
-    The Chess-Pre-trained-Transformer.
+    The **C**hess-**P**re-trained-**T**ransformer.
 
     The rules are simple: provide a PGN string of your current game, the engine will predict the next token!
     """)