In this repository we provide pipelines to improve "skills" of large language models (LLMs). Currently we focus on the ability to solve mathematical problems, but you can use our pipelines for many other tasks as well.
Here are some of the things we support.
- Easily convert models between NeMo, vLLM and TensorRT-LLM formats.
- Host the server in any of the above formats to run large-scale synthetic data generation. You can also call Nvidia NIM API or OpenAI API with the same interface, so it's easy to switch from quick prototyping to large-scale slurm jobs.
- Evaluate your models on many popular benchmarks (it's easy to add new benchmarks or customize
existing settings). The following benchmarks are supported out-of-the-box
- Math problem solving: gsm8k, math, amc23, aime24 (and many more)
- Coding skills: human-eval, mbpp
- Chat/instruction following: ifeval, arena-hard
- General knowledge: mmlu (generative)
- Train models using NeMo-Aligner.
- We support other pipelines as well, such as LLM-based dataset decontamination or using LLM-as-a-judge. And it's easy to add new workflows!
To get started, follow the prerequisites and then run ns --help to see all available
commands and their options.
Using our pipelines we created OpenMathInstruct-2 dataset which consists of 14M question-solution pairs (> 600K unique questions), making it nearly eight times larger than the previous largest open-source math reasoning dataset.
The models trained on this dataset achieve strong results on common mathematical benchmarks.
| model | GSM8K | MATH | AMC 2023 | AIME 2024 | Omni-MATH |
| Llama3.1-8B-Instruct | 84.5 | 51.9 | 9/40 | 2/30 | 12.7 |
| OpenMath2-Llama3.1-8B (nemo | HF) | 91.7 | 67.8 | 16/40 | 3/30 | 22.0 |
| + majority@256 | 94.1 | 76.1 | 23/40 | 3/30 | 24.6 |
| Llama3.1-70B-Instruct | 95.1 | 68.0 | 19/40 | 6/30 | 19.0 |
| OpenMath2-Llama3.1-70B (nemo | HF) | 94.9 | 71.9 | 20/40 | 4/30 | 23.1 |
| + majority@256 | 96.0 | 79.6 | 24/40 | 6/30 | 27.6 |
We provide all instructions to fully reproduce our results.
We also provide a convenient tool for visualizing inference and data analysis
| Overview | Inference Page | Analyze Page |
|---|---|---|
 |
 |
 |
If you find our work useful, please consider citing us!
New paper TBD
@article{toshniwal2024openmath,
title = {OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Dataset},
author = {Shubham Toshniwal and Ivan Moshkov and Sean Narenthiran and Daria Gitman and Fei Jia and Igor Gitman},
year = {2024},
journal = {arXiv preprint arXiv: Arxiv-2402.10176}
}Disclaimer: This project is strictly for research purposes, and not an official product from NVIDIA.