Is GPT-4-LLM open source?

Yes — Instruction-Tuning-with-GPT-4/GPT-4-LLM is open source, released under the Apache-2.0 license.

What language is GPT-4-LLM written in?

Instruction-Tuning-with-GPT-4/GPT-4-LLM is primarily written in HTML.

How popular is GPT-4-LLM?

Instruction-Tuning-with-GPT-4/GPT-4-LLM has 4.3k stars on GitHub.

Where can I find GPT-4-LLM?

Instruction-Tuning-with-GPT-4/GPT-4-LLM is on GitHub at https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM.

← all repositories

Instruction-Tuning-with-GPT-4/GPT-4-LLM

The dataset that lets LLaMA copy GPT-4's homework

It releases GPT-4-generated instruction data so researchers can fine-tune open models without burning API credits.

★4.3k stars HTML Data Tooling Language Models

View on GitHub ↗ Homepage ↗

Not currently ranked — collecting fresh signals.

star history

What it does This repository is a curated data release for instruction tuning. It contains 52K English and 52K Chinese instruction-following examples generated by GPT-4 using Alpaca prompts, plus 9K answers to unnatural instructions and comparison rankings scored by GPT-4 across multiple models. The intended use is to bootstrap open-source LLMs like LLaMA with training material distilled from a stronger closed model.

The interesting bit The authors treat GPT-4 as both tutor and judge. Beyond generating answers, GPT-4 ranked responses from itself, GPT-3.5, and OPT-IML to create reward-model training data. Human evaluation via Amazon Mechanical Turk—using Anthropic’s helpfulness, honesty, and harmlessness criteria—found that a LLaMA model fine-tuned on this GPT-4 data performed substantially better than one trained on GPT-3 data and even matched GPT-4 itself across all three axes.

Key highlights

52K English and 52K Chinese instruction-following datasets in standard Alpaca JSON format, with outputs generated by GPT-4.
comparison_data.json containing GPT-4 quality rankings of model pairs, useful for RLHF reward modeling.
Human evaluation results indicating LLaMA-GPT-4 closes the behavior gap with native GPT-4 on three safety-and-quality criteria.
An included IPython notebook to reproduce the paper’s figure plots from cached results.

Caveats

The data carries a CC BY-NC 4.0 license, meaning research-only, non-commercial use; models trained on it are expected to stay in the lab.
The repo is fundamentally a dataset and plotting utilities; training code and model weights are borrowed from external projects like Alpaca and Vicuna.

Verdict Worth downloading if you’re building or benchmarking instruction-tuned open models. Look elsewhere if you need a commercial-grade dataset or a one-click training pipeline.

Frequently asked

What is Instruction-Tuning-with-GPT-4/GPT-4-LLM?: It releases GPT-4-generated instruction data so researchers can fine-tune open models without burning API credits.
Is GPT-4-LLM open source?: Yes — Instruction-Tuning-with-GPT-4/GPT-4-LLM is open source, released under the Apache-2.0 license.
What language is GPT-4-LLM written in?: Instruction-Tuning-with-GPT-4/GPT-4-LLM is primarily written in HTML.
How popular is GPT-4-LLM?: Instruction-Tuning-with-GPT-4/GPT-4-LLM has 4.3k stars on GitHub.
Where can I find GPT-4-LLM?: Instruction-Tuning-with-GPT-4/GPT-4-LLM is on GitHub at https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM.