import os # os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3,4,5,6,7" os.environ["CUDA_VISIBLE_DEVICES"] = "0" import torch import numpy as np from transformers import AutoModelForSequenceClassification, AutoModelForMultipleChoice, AutoModel from transformers import TrainingArguments, Trainer from transformers import AutoTokenizer from transformers import DataCollatorWithPadding from datasets import load_dataset from datasets import load_metric import torchsnooper model_name = "THUDM/chatglm-6b" if __name__ == "__main__": device = "cuda" if torch.cuda.is_available() else "cpu" tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) model = AutoModel.from_pretrained(model_name, trust_remote_code=True).half().cuda() response, history = model.chat(tokenizer, "你好", history=[]) print(response) response, history = model.chat(tokenizer, "晚上睡不着应该怎么办", history=history) print(response)