Metraipolis › archive › dc_cc0756a47ec6
Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation — arXiv:2609.16145
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.16145
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T18:31:58.251Z
- SHA-256 of the archived text
46325e43e146b055a0968a4b034c4db737123570b8f7d60ebd17af2abd1ad0c3- Stable URL of this page
https://metraipolis.com/documento/dc_cc0756a47ec6
How to cite
Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation — arXiv:2609.16145 — arXiv (en), https://arxiv.org/abs/2609.16145 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cc0756a47ec6 (Metraipolis archive dc_cc0756a47ec6, sha256 46325e43e146b055a0968a4b034c4db737123570b8f7d60ebd17af2abd1ad0c3)
BibTeX
@misc{dccc0756a47ec6,
title = {Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation — arXiv:2609.16145},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.16145},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cc0756a47ec6; sha256 46325e43e146b055a0968a4b034c4db737123570b8f7d60ebd17af2abd1ad0c3},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.16145. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We study a practical question: can a small correction module fix errors in a frozen language model's outputs without degrading its base capabilities? We propose CRN v2, a lightweight logit-level correction module (~34M trainable parameters, 0.73% of the 4.65B text module) that sits atop a fully frozen Gemma 4 E2B model. The base model is never updated; only the correction module learns, via supervised fine-tuning followed by reference-free DPO on 83,400 error-correction pairs. On a 60-question domain exam (CEHRI: Certified Human-Robot Intelligence, covering facts, arithmetic, and implicit-goal reasoning), CRN v2 corrects 53.3% of base-model errors (reworded variant: 43.3%) while showing no degradation on tested capability benchmarks (MMLU/BoolQ N=200; car-wash N=8). A LoRA baseline at the matched CRN v1 budget (6.6M params, rank 19) achieves 83.3% correction but suffers 30-75% capability loss on the same benchmarks -- the correction-capability tradeoff. An ablation shows that the KL preservation term (lambda=0.1) is critical: lowering it to 0.01 degrades correction to 35.0%. A hidden-state injection variant at earlier layers (1.6M params, SFT-only) reaches 50.0%/55.8% but does not exceed logit correction; shallower injection (layer 4) drops to 30.0%/28.3%; multi-depth logit correction (~35M) reaches only 40%; and longer training (5,000 SFT + 2,000 DPO) stays at 53.3% -- none of the alternative configurations we tested exceeded the rank-128 logit result, consistent with a best-achieved result of ~53% rather than a floor. This is a study of a design principle (frozen base + logit correction + KL anchoring), not a claim of architectural novelty. All code, main-result weights, and evaluation scripts are released (deep variant as code only -- no trained deep checkpoints). Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-16 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE). Pagina del lavoro: https://arxiv.org/abs/2609.16145 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-15T18:03:09.409Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-16?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).