Buckets:
| cff-version: 1.2.0 | |
| message: "If you use this work, please cite it as below." | |
| title: "Assumed infrastructure: what do instruction-tuned LLMs take for granted? A pre-registered audit of four small open models" | |
| type: software | |
| authors: | |
| - family-names: "Savadogo" | |
| given-names: "Modeste" | |
| repository-code: "https://github.com/modestesavadogo/assumed-infrastructure" | |
| url: "https://huggingface.co/datasets/SavadogoModeste/assumed-infrastructure-audit" | |
| license: | |
| - MIT | |
| - CC-BY-4.0 | |
| version: 1.0.0 | |
| keywords: | |
| - large language models | |
| - cultural bias | |
| - evaluation | |
| - infrastructure | |
| - pre-registration | |
| - audit | |
| abstract: >- | |
| Four small instruction-tuned chat models (Qwen3-4B-Instruct-2507, Phi-4-mini-instruct, Llama-3.2-3B-Instruct, | |
| Qwen3-1.7B) answer practical-advice requests under ten context conditions. The study measures, with a frozen keyword | |
| lexicon, next-token probes and a blind manual audit, whether the models assume steady electricity, fast internet and | |
| international cards, and how deeply they adapt to a stated place or explicit constraints. Code, inputs and audit | |
| materials are MIT / CC BY 4.0; raw outputs are in the companion Hugging Face dataset. | |
Xet Storage Details
- Size:
- 1.18 kB
- Xet hash:
- d1f072c925eea996c1060bc67aab12d5c31a175a7ecb2a9b6430ec7b2e1516ef
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.