9、生成式AI模型微调与评估：从全量微调到参数高效微调

生成式AI模型微调与评估：从全量微调到参数高效微调

1. 模型微调代码示例

以下是一个来自 train.py 的代码片段，展示了如何使用Hugging Face估计器进行模型微调：

from transformers import (
    AutoModelForCausalLM,
    Trainer,
    TrainingArguments,
)
from datasets import load_from_disk
# Load dataset and convert each row to an instruction prompt
dataset = load_from_disk(...)
dataset = dataset.map(convert_row_to_instruction)
# Define and load the model for fine-tuning
model_checkpoint = "..." # generative model like Llama2, Falcon
model = AutoModelForCausalLM.from_pretrained(model_checkpoint)
# Convert text into tokens using the model's tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
tokenized_dataset = dataset.map(
    lambda row: tokenizer(...) 
)
trai