{"paper":{"arxiv_id":"1909.11942","title":"ALBERT: A Lite BERT for Self-supervised Learning of Language Representations","abstract":"Increasing model size when pretraining natural language representations often results in improved performance on downstream tasks. However, at some point further model increases become harder due to GPU/TPU memory limitations and longer training times. To address these problems, we present two parameter-reduction techniques to lower memory consumption and increase the training speed of BERT. Comprehensive empirical evidence shows that our proposed methods lead to models that scale much better compared to the original BERT.","primary_category":"cs.CL","venue":"ICLR 2020","published_at":null,"latest_version":1,"withdrawn":false},"latest_version":{"id":"fe1b61ed-9d9d-4c6c-93a2-e3768963a2b1","version":1,"source_url":"https://arxiv.org/abs/1909.11942","rendered_html_url":null,"rendering_engine":null},"verdict":{"id":"954925a1-6e93-4e48-a934-a103fb3628e7","kind":"POST","status":"reproduced","score":0.9066666666666667,"confidence":0.8,"agent_version":"v0.1.0-albert-mrpc-microslice","computed_at":"2026-05-14T23:17:42.537Z","is_current":true,"claim_citation":{"paper_arxiv_id":"1909.11942","section":"Table 2","row":"ALBERT-base","column":"MRPC F1","reported_value":89.3,"reported_metric":"f1","quoted_text":"ALBERT-base 89.3","pdf_page":7,"notes":"Table 2 (Dev set results on the GLUE benchmark) of arXiv:1909.11942. ALBERT-base MRPC F1 = 89.3. The matching community checkpoint is `textattack/albert-base-v2-MRPC` (v2 release; original paper used v1). Driver evaluates an MRPC dev micro-slice, so PROTOCOL_MATCH is `proxy`."},"protocol_match":"proxy"},"verdicts":{"post":{"id":"954925a1-6e93-4e48-a934-a103fb3628e7","kind":"POST","status":"reproduced","score":0.9066666666666667,"confidence":0.8,"agent_version":"v0.1.0-albert-mrpc-microslice","computed_at":"2026-05-14T23:17:42.537Z","is_current":true,"claim_citation":{"paper_arxiv_id":"1909.11942","section":"Table 2","row":"ALBERT-base","column":"MRPC F1","reported_value":89.3,"reported_metric":"f1","quoted_text":"ALBERT-base 89.3","pdf_page":7,"notes":"Table 2 (Dev set results on the GLUE benchmark) of arXiv:1909.11942. ALBERT-base MRPC F1 = 89.3. The matching community checkpoint is `textattack/albert-base-v2-MRPC` (v2 release; original paper used v1). Driver evaluates an MRPC dev micro-slice, so PROTOCOL_MATCH is `proxy`."},"protocol_match":"proxy"},"pre":null}}