因果语言模型

1、导入相关包

from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, DataCollatorForLanguageModeling, TrainingArguments, Trainer, BloomForCausalLM

2、加载、划分、处理数据集

模型地址 国内链接

# 加载
ds = Dataset.load_from_disk("./wiki_cn_filtered/")
# 划分

# 数据处理
tokenizer = AutoTokenizer.from_pretrained("./bloom-389m-zh")

def process_func(examples):
    contents = [e + tokenizer.eos_token for e in examples["completion"]]
    return tokenizer(contents, max_length=384, truncation=True)

tokenized_ds = ds.map(process_func, batched=True, remove_columns=ds.column_names)  

3、创建模型

model = AutoModelForCausalLM.from_pretrained("./bloom-389m-zh")

4、创建评估函数

5、创建TrainingArguments、Trainer

args = TrainingArguments(
    output_dir="./causal_lm",
    per_device_train_batch_size=2,
    gradient_accumulation_steps=16,
    logging_steps=10,
    num_train_epochs=1,
    fp16=True
)

trainer = Trainer(
    args=args,
    model=model,
    tokenizer=tokenizer,
    train_dataset=tokenized_ds,
    data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False)
)

6、模型训练、评估、预测

# 模型训练
trainer.train()
# 模型评估
trainer.evaluate()

7、模型预测

# 模型预测

# 测试
from transformers import pipeline

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)

pipe("西安交通大学博物馆（Xi'an Jiaotong University Museum）是一座位于西安", max_length=128, do_sample=True)

pipe("下面是一则游戏新闻。小编报道，近日，游戏产业发展的非常", max_length=128, do_sample=True)