Open Instruct V1
A textual corpus that compiles several open-source data sets to train language models to understand and execute instructions, ideal for fine-tuning conversational agents.
Approximately 222,650 examples, structured text format (instruction/output), JSON
Apache 2.0
Description
The dataset Open Instruct V1 is a corpus of approximately 222,000 instruction-response pairs, gathered from a variety of sources such as Alpaca, Dolly, GPT-4, Self-Instruct, and others. All entries have been cleaned and standardized in a single format to make it easier to train language models to follow instructions in a consistent and relevant manner.
What is this dataset for?
- Train LLM models to follow textual instructions in a structured way
- Refine a pre-trained model for interactive tasks (QA, dialogue, virtual assistance)
- Create chatbots that can interpret complex queries
Can it be enriched or improved?
Yes, the dataset is modular and extensible. It is possible to add new instruction-response pairs adapted to a specific field, to vary the styles (e.g. formal/informal tone), or to filter the instructions according to their complexity to refine the training.
🔎 In summary
🧠 Recommended for
- Generative AI researchers
- Conversational assistant designers
- LoRa experimenters
🔧 Compatible tools
- Hugging Face Transformers
- Axolotl
- OpenChatKit
- VLLM
💡 Tip
Filter by type of instruction (question, dialog, code) to better target uses in fine-tuning.
Frequently Asked Questions
What types of tasks can you train with this dataset?
This dataset allows you to train models to follow instructions, generate consistent answers, perform question-answer or reformulation.
Is it necessary to rework the data before use?
No, the data is already cleaned and formatted consistently. They are ready to use for fine tuning.
Is this dataset suitable for multilingual use?
It is mostly in English, but can be extended to other languages by translating or adding new multi-lingual instructions.
