to cpu

app.py

@@ -11,7 +11,14 @@ import gradio as gr
 from transformers import AutoTokenizer, GenerationConfig, AutoModel
 
 
-model = AutoModel.from_pretrained("KumaTea/twitter-int8", trust_remote_code=True, revision="693e7b1").float()
+device = torch.device('cpu')
+torch.cuda.current_device = lambda : device
+
+model = AutoModel.from_pretrained(
+    "KumaTea/twitter-int8",
+    trust_remote_code=True,
+    revision="693e7b1"
+).float().to(device)
 tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True, revision="4de8efe")
 
 # dump a log to ensure everything works well