OpenELM 3B Instruct
Release in the OpenELM family · version OpenELM-3B-Instruct
OpenELM 3B Instruct is the 3B-parameter instruction-tuned model in Apple's April 2024 OpenELM release. It was pretrained with Apple's CoreNet library on about 1.8 trillion tokens of public data and then instruction-tuned on the UltraFeedback dataset using the Hugging Face Alignment Handbook.17
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Repository: CoreNet repository (external site: github.com)
- Paper: OpenELM paper (arXiv 2404.14619) (external site: arxiv.org)
- License: License (Apple Machine Learning Research Model License) (external site: huggingface.co)
Availability and license
Overall availability
Downloadable from Hugging Face without an access gate, under a license that permits only non-commercial research use.23
Availability is separate from permission: read the license before using or redistributing.
Apple Machine Learning Research Model License Agreement (external site: huggingface.co)32
Apple software license (CoreNet) (external site: raw.githubusercontent.com)4
The weights license grants a personal, non-exclusive, revocable license to use, copy, modify, distribute, and create derivatives only for non-commercial scientific research, excluding product development and commercial use; derivatives are held to the same limit. Redistribution requires passing on the agreement with a specified attribution notice and identifying modifications. No patent rights are granted, and the agreement is governed by California law. The card's metadata uses the license tag "apple-amlr" but still gives "apple-sample-code-license" as the license name, while the LICENSE file in the repository (copyright 2025) is the research-only agreement. CoreNet's own license is a permissive Apple software license that grants no patent rights.324
Model-disclosure tier
Open-weight, plus published inference code, training code, and training recipe, and at least documented training-data composition.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Weights are published on Hugging Face without an access gate, under a research-only license.23 |
| Inference codeIs code for running the model published? | Public | The Hugging Face repository includes model code and a generate_openelm.py script for inference with Transformers; CoreNet documents conversion to Apple's MLX.125 |
| Training codeIs the code used to train the model published? | Public | CoreNet documents OpenELM pre-training, and the instruction-tuning recipe uses the public Alignment Handbook with a configuration published in CoreNet.567 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | Pre-training used public datasets (RefinedWeb, deduplicated PILE, a subset of RedPajama, and a subset of Dolma v1.6, about 1.8T tokens), and CoreNet documents where to download them and how the subsets were specified; instruction tuning used UltraFeedback. Current availability of each source dataset was not verified.167 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Public | Apple published pre-training configurations and training logs, and the instruction-tuning README gives the recipe file and per-size hyperparameters.87 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Public | The model card gives the lm-evaluation-harness commit, tasks, and shot settings used to reproduce its reported zero-shot and few-shot results.1 |
What it is useful for
Run and use notes
Organization context
Provenance and derivatives
Pretrained from scratch by Apple on public datasets using CoreNet, then instruction-tuned by Apple. The models use Meta's Llama tokenizer, which is not distributed with them.157
- Derived from: OpenELM-3B (pretrained) (external site: huggingface.co) — Pretrained base checkpoint; no separate catalog record.
Other releases in the OpenELM family
No other releases in this family have been assessed.
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.