zhllama1.0带带缝合机（4Bit量化）可以在8G显存下流畅推理，喂了
统一alpaca数据集训练，
感谢unsloth开源框架支持，感谢谷歌colab免费T4无私奉献
该模型仅为适配器，请挂载
base_model, base_tokenizer = FastLanguageModel.from_pretrained(
model_name="unsloth/meta-llama-3.1-8b-bnb-4bit", # 基础模型名称
max_seq_length=max_seq_length,
dtype=dtype,
load_in_4bit=load_in_4bit,
)
model = PeftModel.from_pretrained(base_model, "DudeGuuud/zh_llama3_1.0", load_in_4bit=load_in_4bit) # 替换为你的适配器名称
tokenizer = base_tokenizer # 分词器与基础模型相同
This llama model was trained 2x faster with
Unsloth and Huggingface's TRL library.
