samsum_instruction_tune

Instruction-tuning version of the SAMSum dialogue summarization dataset, formatted for training large language models (LLMs) with instruction-following capabilities.

---Overview

Atmanstr/samsum_instruction_tune is a reformatted version of the SAMSum dataset designed specifically for instruction tuning. Each example converts the original dialogue summarization task into an instruction–response format suitable for fine-tuning chat-based or instruction-following models.

This dataset helps train models to:

Understand multi-turn dialogues

Extract key information from conversations

Generate concise summaries

Follow natural language instructions

Dataset Source

---The dataset is derived from:

SAMSum: Abstractive Dialogue Summarization Dataset

Created by: Gliwa et al.

Original dataset contains conversations resembling real-world messenger chats between two or more people.

---Format

Each dataset entry follows an instruction tuning format:

{ "instruction": "Summarize the following conversation.", "input": "", "output": "

" } Example { "instruction": "Summarize the following conversation.", "input": "Amanda: Are we still meeting today?\nJerry: Yes, at 5 PM.\nAmanda: Great, see you there!", "output": "Amanda and Jerry confirm their meeting at 5 PM." }

---This dataset can be used for:

Instruction tuning LLMs

Dialogue summarization tasks

Fine-tuning chat models

Benchmarking conversational understanding

Common model families:

LLaMA / LLaMA 2 / LLaMA 3

Mistral

Falcon

GPT-style models

T5 / FLAN-T5

You can format samples for supervised fine-tuning like:

Instruction:

Summarize the following conversation.

Input:

Response:

This format works well for:

LoRA fine-tuning

QLoRA training

Supervised Fine-Tuning (SFT)