BWhat changed
Released a layer-scaled language-model family with public-data preparation, training recipes, checkpoints, logs and evaluation framework, plus conversion paths for Apple-device inference and fine-tuning.
Scientific scope
Released a layer-scaled language-model family with public-data preparation, training recipes, checkpoints, logs and evaluation framework, plus conversion paths for Apple-device inference and fine-tuning. Layer-wise scaling is explicitly inherited from DeLighT; GQA, RoPE, SwiGLU and FlashAttention also retain prior credit. Acquired contribution is the integrated reproducible training resource, not invention of those primitives. Reported efficiency comparisons depend on training budgets and benchmark choices.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Layer-wise scaling is explicitly inherited from DeLighT; GQA, RoPE, SwiGLU and FlashAttention also retain prior credit. Acquired contribution is the integrated reproducible training resource, not invention of those primitives. Reported efficiency comparisons depend on training budgets and benchmark choices.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- OpenELM: An Efficient Language Model Family with Open-source Training and Inference Framework
- Mehta et al.: OpenELM reproducible layer-scaled language-model framework — official research provenance
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction