Dataset · nvidia
5M instructions for coding LLMs
nvidia/OpenCodeInstruct
5 million examples for SFT of coding LLMs: diverse English instructions from NVIDIA.
What's inside
The dataset contains 5 million diverse examples for supervised fine-tuning (SFT) of coding LLMs. According to the authors, it is the largest open instruction tuning dataset; the language is English. The card includes a technical report, a GitHub pipeline for SFT, and a note that the data is ready for use.
What you can do with it
- Fine-tuning a coding LLM via SFT
- Evaluating instruction following in code generation
- Comparing the quality of coding models after SFT
- Tasks
- Text generation
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 59.1K
- Likes
- 121
Models trained on it
- mradermacher/Viper-9B-i1-GGUF 3.2K
- mradermacher/Frank-26B-A4B-i1-GGUF 2.3K
- Ternii/Qwen3.6-35B-A3B-TR2.25-G-GGUFText generation 1.6K
- mradermacher/Viper-9B-GGUF 1.4K
- mradermacher/Frank-26B-A4B-GGUF 1.1K
- mradermacher/Atomight-V2.5-1.7B-C1-i1-GGUF 1K
- TendieLabs/Frank-26B-A4B-GGUFsText generation 845
- Ternii/Qwen3.8-Flash-Next-TR2.31-G-GGUFText generation 621
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/OpenCodeInstruct")