{
  "id": "2402.12264",
  "version": "v1",
  "published": "2024-02-19T16:26:00.000Z",
  "updated": "2024-02-19T16:26:00.000Z",
  "title": "Uncertainty quantification in fine-tuned LLMs using LoRA ensembles",
  "authors": [
    "Oleksandr Balabanov",
    "Hampus Linander"
  ],
  "comment": "8 pages, 4 figures",
  "categories": [
    "cs.LG",
    "cs.AI",
    "cs.CL",
    "stat.ML"
  ],
  "abstract": "Fine-tuning large language models can improve task specific performance, although a general understanding of what the fine-tuned model has learned, forgotten and how to trust its predictions is still missing. We derive principled uncertainty quantification for fine-tuned LLMs with posterior approximations using computationally efficient low-rank adaptation ensembles. We analyze three common multiple-choice datasets using low-rank adaptation ensembles based on Mistral-7b, and draw quantitative and qualitative conclusions on their perceived complexity and model efficacy on the different target domains during and after fine-tuning. In particular, backed by the numerical experiments, we hypothesise about signals from entropic uncertainty measures for data domains that are inherently difficult for a given architecture to learn.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2024-02-19T16:26:00.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "uncertainty quantification",
      "fine-tuned llms",
      "lora ensembles",
      "computationally efficient low-rank adaptation ensembles",
      "fine-tuning large language models"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 8,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}