Awesome-instruction-tuning
收藏资源简介:
这是一个精选的数据集合集,专注于指令调优领域,收集和整理了多个开源指令调优数据集。合集覆盖了从传统NLP任务修改而来的数据集以及由大型语言模型生成的数据集,包括UnifiedQA、CrossFit、Natural Instructions、Flan、P3、MetaICL、ExMix、Super-Natural Instructions、GLM、xP3、Unnatural Instructions、Alpaca、Dolly、Guanaco、Chinese-Vicuna等。内容以表格形式组织,提供数据集发布年份、名称、任务数量、实例数量、相关模型等详细信息,旨在为研究人员和开发者提供全面的指令调优数据集资源索引。
This is a curated dataset collection focused on the instruction tuning domain, which collects and organizes multiple open-source instruction tuning datasets. The collection covers datasets adapted from traditional NLP tasks and datasets generated by large language models, including UnifiedQA, CrossFit, Natural Instructions, Flan, P3, MetaICL, ExMix, Super-Natural Instructions, GLM, xP3, Unnatural Instructions, Alpaca, Dolly, Guanaco, Chinese-Vicuna, and more. The content is structured in tabular format, providing detailed information such as the dataset's release year, name, number of tasks, number of instances, and associated models, aiming to offer a comprehensive resource index of instruction tuning datasets for researchers and developers.
数据集详情总结
该页面是一个关于指令微调(Instruction Tuning)的开源资源汇总仓库,涵盖了数据集、模型、论文和相关仓库。
数据集与模型
1. 基于传统NLP任务改造的数据集
以下数据集由传统NLP任务改造而来,按发布时间排序:
| 发布时间 | 数据集 | 任务数量 | 实例数量 | 对应模型 | 基座模型 | 模型规模 |
|---|---|---|---|---|---|---|
| 2020-05 | UnifiedQA | 46 | 750k | UnifiedQA | RoBerta | 110-340 M |
| 2021-04 | CrossFit | 159 | 71M | BART-CrossFit | BART | 140 M |
| 2021-04 | Natural Inst v1.0 | 61 | 620k | Gen. BART | BART | 140 M |
| 2021-09 | Flan 2021 | 62 | 4.4M | Flan-LaMDA | LaMDA | 137B |
| 2021-10 | P3 | 62 | 12M | TO, TO+, TO++ | T5-LM | 3-11B |
| 2021-10 | MetalCL | 142 | 3.5M | MetalCL | GPT-2 | 770 M |
| 2021-11 | ExMix | 107 | 500k | ExT5 | T5 | 220M-11B |
| 2022-04 | Super-Natural Inst. | 1613 | 5M | Tk-Instruct | T5-LM, mT5 | 17-13B |
| 2022-10 | GLM | 77 | 12M | GLM-130B | GLM | 130 B |
| 2022-10 | Flan 2022 | 1836 | 15M | Flan-T5, Flan-PaLM | T5-LM, PaLM | 10 M-540 B |
| 2022-11 | xP3 | 71 | 81M | BLOOMz, mTO | BLOOM, mT5 | 13-176B |
| 2022-12 | Unnatural Inst. | 117 | 64k | T5-LM-Unnat. Inst. | T5-LM | 11B |
2. 由大语言模型生成的数据集
| 发布时间 | 模型名称 | 基座模型 | 模型规模 | 数据集 | 实例数量 | 语言 |
|---|---|---|---|---|---|---|
| 2022-12 | GPT-3 Self Inst. | GPT-3 | 175B | Self-Instruct | 82k | 英文 |
| 2023-03-03 | alpaca | LLaMA | 7B | alpaca_data | 52k | 英文 |
| 2023-03-19 | alpaca-lora | LLaMA | 7B, 13B, 30B | alpaca_data, alpaca_data_cleaned | 52k | 英文 |
| 2023-03-23 | Chinese-Vicuna | LLaMA | 7B, 13B | BELLE, GuanacoDataset | 1M | 中文 |
| 2023-03-24 | Alpaca-CoT | LLaMA | 7B | dataset | — | 中英文 |
| 2023-03-25 | dolly | dolly | 6B | alpaca_data | 52k | 英文 |
| 2023-03-25 | guanaco | LLaMA | 7B | GuanacoDataset | 534k | 英/中/日/德 |
| 2023-03-28 | Chinese-LLaMA-Alpaca | LLaMA | 7B | alpaca_data_zh, pCLUE, translation2019zh, alpaca_data, Self-Instruct | 2M | 中文 |
| 2023-03-29 | ColossalChat | LLaMA | 7B, 13B | InstructionWild | 104k | 中英文 |
| 2023-03-31 | Luotuo | LLaMA, ChatGLM | 7B, 6B | trans_chinese_alpaca_data | 52k | 中文 |
| 2023-03-31 | cerebras-lora-alpaca | Cerebras-GPT | 2.7B | AlpacaDataCleaned | 52k | 英文 |
多语言翻译工具
页面提供了基于 Helsinki-NLP 的开源翻译工具,可将英文数据集翻译为超过100种语言。
- translator.py:用于翻译数据集,支持各种 Helsinki-NLP 模型
- 用法示例:
python translator.py Helsinki-NLP/opus-mt-en-zh alpaca_data.json
- 用法示例:
- process.py:用于处理翻译结果,可删除包含连续重复三次及以上字符串的提示
- 用法示例:
python process.py translated_data.json
- 用法示例:
- 提供了最终处理后的翻译数据文件:
translated_alpaca_data.json - 对于超过最大输入长度的提示,会在翻译前丢弃
相关论文
页面列出了以下具有代表性的论文(按时间排序):
- Finetuned language models are zero-shot learners (2021.9)
- Multitask Prompted Training Enables Zero-Shot Task Generalization (2021.10)
- Training language models to follow instructions with human feedback (2022.3)
- Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks (2022.4)
- Unsupervised Cross-Task Generalization via Retrieval Augmentation (2022.4)
- Instruction Induction: From Few Examples to Natural Language Task Descriptions (2022.5)
- Scaling Instruction-Finetuned Language Models (2022.10)
- Guess the Instruction! Flipped Learning Makes Language Models Stronger Zero-Shot Learners (2022.10)
- Unnatural Instructions: Tuning Language Models with (Almost) No Human Labor (2022.12)
- Improving Cross-task Generalization of Unified Table-to-text Models with Compositional Task Configurations (2022.12)
- Self-Instruct: Aligning Language Model with Self Generated Instructions (2022.12)
- MultiInstruct: Improving Multi-Modal Zero-Shot Learning via Instruction Tuning (2022.12)
- The Flan Collection: Designing Data and Methods for Effective Instruction Tuning (2023.1)
- In-Context Instruction Learning (2023.2)
相关仓库
指令学习
- awesome-instruction-learning
- awesome-instruction-dataset
上下文学习(ICL)
- ICL_PaperList
- prompt-in-context-learning
推理
- LM-reasoning
- LLM-Reasoning-Papers
- Chain-of-ThoughtsPapers
框架
- OpenICL




