fix gpu version

app.py

@@ -13,19 +13,21 @@ import torch
 #                                      else lora_folder),
 #                                     trust_remote_code=True)
 # model = AutoModelForCausalLM.from_pretrained(("baichuan-inc/Baichuan-13B-Base" if model_folder == ''
-#                                               else model_folder), torch_dtype=torch.float32, trust_remote_code=True)
+#                                               else model_folder),
+#                                               torch_dtype=torch.float16,
+#                                               device_map="auto",
+#                                               trust_remote_code=True)
 # model = PeftModel.from_pretrained(model,
 #                                   ("Junity/Genshin-World-Model" if lora_folder == ''
-#                                    else lora_folder)
-#                                   , torch_dtype=torch.float32, trust_remote_code=True)
+#                                    else lora_folder),
+#                                    device_map="auto",
+#                                    torch_dtype=torch.float32,
+#                                    trust_remote_code=True)
 # tokenizer = AutoTokenizer.from_pretrained(("baichuan-inc/Baichuan-13B-Base" if model_folder == ''
 #                                            else model_folder),
 #                                           trust_remote_code=True)
-# history = []
-# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-# if device == "cuda":
-#     model.cuda()
-#     model = model.half()
+history = []
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
 
 def respond(role_name, character_name, msg, textbox, temp, rep, max_len, top_p, top_k):

requirements.txt

@@ -1,7 +1,7 @@
 gradio==3.40.1
 peft==0.4.0
-torch==2.0.0+cu118
-transformers==4.31.0
-transformers_stream_generator==0.0.4
-tiktoken
-sentencepiece
+transformers_stream_generator
+sentencepiece
+accelerate
+colorama
+cpm_kernels