Implementation:LLMBook zh LLMBook zh github io SFTDataset Format Template

Knowledge Sources	LLMBook-zh
Domains	NLP, Data_Engineering
Last Updated	2026-02-08 00:00 GMT

Overview

Concrete tool for Alpaca-style instruction template formatting provided by the LLMBook repository.

Description

The SFTDataset.format_template class attribute is a dictionary containing two prompt templates: prompt_input (for instructions with additional input) and prompt_no_input (for standalone instructions). These templates use Python string formatting placeholders ({instruction}, {input}) to construct the full prompt string.

Usage

This is used internally by SFTDataset.process() to format each training example before tokenization.

Code Reference

Source Location

• Repository: LLMBook-zh
• File: code/7.2 SFT数据类.py
• Lines: 9-20 (template definitions); 48-57 (usage in process)

Signature

class SFTDataset:
    instruction_template = "\n### Instruction:\n"
    response_template = "\n### Output:\n"
    format_template = {
        "prompt_input": str,    # Template with {instruction} and {input} placeholders
        "prompt_no_input": str, # Template with {instruction} placeholder only
    }

Import

from dataset.sft_dataset import SFTDataset

I/O Contract

Inputs

Name	Type	Required	Description
instruction	str	Yes	The task instruction text
input	str	No	Additional input context (determines which template is used)
output/response	str	Yes	The expected response text

Outputs

Name	Type	Description
formatted_prompt	str	Complete formatted prompt string with template markers

Usage Examples

from dataset.sft_dataset import SFTDataset

# Example with input
example = {"instruction": "Summarize the text.", "input": "The cat sat on the mat.", "output": "A cat rests on a mat."}
prompt = SFTDataset.format_template["prompt_input"].format_map(example)

# Example without input
example2 = {"instruction": "Tell me a joke."}
prompt2 = SFTDataset.format_template["prompt_no_input"].format_map(example2)

Related Pages

Implements Principle

• Principle:LLMBook_zh_LLMBook_zh_github_io_Instruction_Data_Formatting