[
    {
        "id": "osp-15461",
        "type": "article-journal",
        "title": "Algorithmic Scratchpads and Curriculum Staging for Arithmetic Reasoning in Tiny Transformers",
        "author": [
            {
                "family": "Kasliwal",
                "given": "Sourabh"
            }
        ],
        "URL": "https://omanscience.com/en/articles/algorithmic-scratchpads-and-curriculum-staging-for-arithmetic-reasoning-in-tiny-transformers",
        "language": "en",
        "issued": {
            "date-parts": [
                [
                    2026
                ]
            ]
        },
        "abstract": "Autoregressive Large Language Models (LLMs) frequently struggle with deterministic multi-step algorithmic tasks such as multi-digit multiplication and long division. In this paper, we investigate the mechanics of multi-step arithmetic in compact \"Tiny\" Transformers (~10.6M non-embedding parameters, 49.3M total) trained on synthetic data across four basic operations (+, -, *, /) unrolled as step-by-step scratchpads. First, we establish the necessary training foundations: (1) dataloader sequence padding creates an 83% gradient starvation artifact that collapses accuracy from 40% to 1%, remediated via continuous sequence packing; (2) linguistic pretraining is an essential prerequisite ("
    }
]