from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel,PeftConfig
# 例如: finetune_model_path='FlagAlpha/Llama2-Chinese-7b-Chat-LoRA'
config = PeftConfig.from_pretrained(finetune_model_path)
# 例如: base_model_name_or_path='meta-llama/Llama-2-7b-chat'
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path,use_fast=False)
tokenizer.pad_token = tokenizer.eos_token
device_map = "cuda:0" if torch.cuda.is_available() else "auto"
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path,device_map=device_map,torch_dtype=torch.float16,load_in_8bit=True,trust_remote_code=True,use_flash_attention_2=True)
model = PeftModel.from_pretrained(model, finetune_model_path, device_map={"": 0})
input_ids = tokenizer(['<s>Human: 介绍一下北京\n</s><s>Assistant: '], return_tensors="pt",add_special_tokens=False).input_ids
if torch.cuda.is_available():
input_ids = input_ids.to('cuda')
"repetition_penalty":1.3,
"eos_token_id":tokenizer.eos_token_id,
"bos_token_id":tokenizer.bos_token_id,
"pad_token_id":tokenizer.pad_token_id
generate_ids = model.generate(**generate_input)
text = tokenizer.decode(generate_ids[0])