Nothing Changed but the Model: CellFill -- Bounded In-Cell Learning for Bit-Identical, Revocable Updates to Quantized LLMs

• URL: http://arxiv.org/abs/2608.20873v1
• Date: Fri, 21 Aug 2026 08:49:07 GMT
• Title: Nothing Changed but the Model: CellFill -- Bounded In-Cell Learning for Bit-Identical, Revocable Updates to Quantized LLMs
• Abstract summary: We show that every way of teaching a language model, full fine-tuning, adapter merging, model editing, replaces the released checkpoint.<n>With the integer codes and scales of a 4-bit release frozen, new knowledge is written only into the per-weight residual that lives strictly inside each quantization decision cell.<n>We give six propositions and three training paths, including CellFill.
• Score: 4.455667003143275
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Every way of teaching a deployed language model something new -- full fine-tuning, adapter merging, model editing -- replaces the released checkpoint, and with it every evaluation and cache that referred to those exact bits. We instead learn inside the dequantization gap: with the integer codes and scales of a 4-bit release frozen, new knowledge is written only into the per-weight residual that lives strictly inside each quantization decision cell. Re-quantization then returns the released artifact bit-for-bit, a machine-checkable guarantee; updates are exactly revocable by dropping the residual; and drift is bounded. We give six propositions and three training paths, including CellFill, a bounded reparameterization that makes invariance structural rather than enforced. Exact invariance turns out to be nearly free: across three paired seeds the constrained dense path matches an unconstrained reference whose weights provably escape the artifact (58.9 vs 59.3 percent fact recall; paired difference -0.5 points, 95% CI [-5.0,+4.0]), and is better on held-out cross-domain perplexity. Against the natural null hypothesis -- serving the same update as an unmerged adapter -- projecting into the cells reduces cross-domain forgetting in every run that converged, and a diverged control shows the boundary: projection is a trust region, not a repair. What no method escapes is the cost of knowledge itself, and the apparent free lunch of in-domain perplexity improving past the anchor is an artifact of rehearsal sharing a corpus with the metric. Methods differ threefold at matched rehearsal in knowledge bought per point of cross-domain perplexity, a ranking that is not the recall ranking. The method transfers to a 27B hybrid linear-attention model (2.4e10 constrained weights, verified bit-identical), where matched recall costs about half as much cross-domain perplexity as at 1.7B.

Related papers

• NovaFabric: Tamper-Evident, Replayable Evidence for Autonomous AI Agent Runs [0.0]
  We present NovaFabric, producing audit-grade execution evidence.<n>It records an agent run, without modifying agent logic, into a portable Run Capsule.<n>Sealed runs are re-executable under a four-mode replay protocol.<n>We evaluate eight research questions at measured scope.
  arXiv  Detail & Related papers  (2026-09-11T08:35:43Z)
• Tied Trit-Planes: Constraining PTQTP to a Uniform Nine-Level Quantizer, with a Persistent Folded Format for Disk-Streamed Mixture-of-Experts Serving [0.28304882054363123]
  This work is the first to impose that identity as a constraint inside PTQTP's solver.<n>We apply this to the routed experts of DeepSeek-V4-Flash-0731, a 284B-A13B mixture-of-experts model.<n>All code, formats, and evaluation artifacts are open source in the fucina inference stack.
  arXiv  Detail & Related papers  (2026-08-09T20:53:47Z)
• Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge Flywheel [0.0]
  We make a frozen small language model both more capable and dramatically cheaper at once, without changing any weights.<n>We show that own-position graft is the numerically exact operating point on a model with floating-point rotary encoding.<n>We describe the system at the behavior level; the engine is proprietary, and every reported number is backed by committed input and output so the scoring can be re-checked without it.
  arXiv  Detail & Related papers  (2026-07-15T23:55:48Z)
• Requential Coding: Pushing the Limits of Model Compression with Self-Generated Training Data [55.04392122319273]
  We introduce requential coding, where a teacher model selects training samples drawn from the student's own distribution.<n>The resulting code length is independent of parameter count and data entropy, and often orders of magnitude shorter than the prequential counterpart.
  arXiv  Detail & Related papers  (2026-07-13T17:58:50Z)
• Forgetting Is Not a Fix: Path Dependence in Sequential Engram Editing [35.76482964927589]
  We test the hypothesis that concept-specific memory traces become linear objects that can be extracted once and combined arithmetically.<n>We run that test on the authors' own reference implementation, at their reported best edit strength.<n>For unlearning-as-compliance: erasure certified today does not certify the artifact after its next edit.
  arXiv  Detail & Related papers  (2026-07-06T23:45:06Z)
• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• Right Makes Might: Aligning Verified Hidden States Empowers RL Reasoning [55.264863369127774]
  Current methods reduce each correct rollout to a single reward bit, ignoring the geometric structure shared among their hidden states.<n>We propose Hidden-Align, an auxiliary loss function that aligns the last-layer hidden states of correct rollouts at the anchor token during RL training, with zero overhead in both training and inference.<n>On eight mathematical reasoning benchmarks, Hidden-Align improves average pass@1 over the DAPO baseline by 3.8, 6.2, and 5.4 percentage points on Qwen3-1.7B, 4B, and 14B respectively, with consistent pass@k gains across all three scales.
  arXiv  Detail & Related papers  (2026-06-02T06:51:15Z)
• CRMA: A Spectrally-Bounded Backbone for Modular Continual Fine-Tuning of LLMs [0.0]
  We introduce CRMA, a residual adapter whose internal mixing matrix M is doubly-stochastic at every forward pass via Sinkhorn normalization.<n>On Mistral-7B across 5 domains 3 seeds, modular per-task LoRA mediates a backbone loss-relative drift from +42.96% +/- 5.5 (naive fine-tuning) to -0.17% +/- 0.17, with disjoint per-seed ranges, and improves prior-task holdout loss by 1.99% +/- 0.54 over a matched frozen-substrate baseline.
  arXiv  Detail & Related papers  (2026-05-29T21:50:23Z)
• CommitDistill: A Lightweight Knowledge-Centric Memory Layer for Software Repositories [0.5872014229110213]
  Software accumulate large amounts of unstructured knowledge in commit messages, pull-request discussions, and issue threads.<n>Recent work on typed-memory architectures for agents (MemGPT, generative agents, and the PlugMem module of Yang et al.) argues that agent memory should be distilled, typed knowledge rather than raw interaction text.<n>We adapt that stance to a software repository's own git history under a constrained regime: deterministic, dependency-free, local-only, no embeddings.
  arXiv  Detail & Related papers  (2026-05-18T12:14:28Z)
• Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution [3.6704226968275253]
  We show gradient-based methods that achieve meaningful forgetting lose it under compression, while methods that survive quantization barely change the model.<n>We present MANSU, which resolves both modes by combining causal circuit attribution to isolate the minimal forget-set subgraph.<n>We additionally introduce Circuit Attribution Divergence (CAD), a mechanistic verification metric distinguishing structural erasure from behavioral suppression.
  arXiv  Detail & Related papers  (2026-05-14T17:44:10Z)
• How Far Is Document Parsing from Solved? PureDocBench: A Source-TraceableBenchmark across Clean, Degraded, and Real-World Settings [56.70440596502351]
  The past year has seen over 20 open document parsing models, yet benchmarks almost exclusively on OmniDocBench.<n>We present PureDocBench, a benchmark thatrenders document images from HTML/CSS, covering 10 domains, 66 pages, and 1,475 pages.
  arXiv  Detail & Related papers  (2026-05-08T09:30:31Z)
• Keeping Code-Aware LLMs Fresh: Full Refresh, In-Context Deltas, and Incremental Fine-Tuning [0.0]
  We study how to keep a model fresh without surrendering retention on earlier code.<n>We compare three families of update strategies: (A) Full Refresh, retraining the entire model at the new snapshot; (B) In-Context Learning (ICL) that injects recent deltas at inference; and (C) Incremental Fine-Tuning (Inc-FT) on delta-derived training sets.<n>We contribute an alias-aware evaluation protocol that credits rename while never rewarding deleted paths, and a practical Forgetting Probe that quantifies residual emissions of obsolete paths.
  arXiv  Detail & Related papers  (2025-11-18T01:01:56Z)
• Continual Learning via Sparse Memory Finetuning [58.163704181154834]
  We investigate whether sparse parameter updates can enable learning without catastrophic forgetting.<n>By updating only the memory slots that are highly activated by a new piece of knowledge relative to usage on pretraining data, we reduce interference between new knowledge and the model's existing capabilities.<n>We find that sparse memory finetuning learns new knowledge while exhibiting substantially less forgetting.
  arXiv  Detail & Related papers  (2025-10-16T19:44:38Z)
• Ensemble Threshold Calibration for Stable Sensitivity Control [0.0]
  We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs.<n>Our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.
  arXiv  Detail & Related papers  (2025-10-02T15:22:28Z)
• Apertus: Democratizing Open and Compliant LLMs for Global Language Environments [163.70368742538187]
  Apertus is a fully open suite of large language models (LLMs) designed to address two systemic shortcomings in today's open model ecosystem.<n>Apertus models are pretrained exclusively on openly available data, respecting robots.txt exclusions and filtering for non-permissive, toxic, and personally identifiable content.<n>The Apertus models also expand multilingual coverage, training on 15T tokens from over 1800 languages, with 40% of pretraining data allocated to non-English content.
  arXiv  Detail & Related papers  (2025-09-17T17:59:21Z)
• Erasing Without Remembering: Implicit Knowledge Forgetting in Large Language Models [81.62767292169225]
  We investigate knowledge forgetting in large language models with a focus on its generalisation.<n>We propose PerMU, a novel probability perturbation-based unlearning paradigm.<n>Experiments are conducted on a diverse range of datasets, including TOFU, Harry Potter, ZsRE, WMDP, and MUSE.
  arXiv  Detail & Related papers  (2025-02-27T11:03:33Z)
• Language Modeling with Editable External Knowledge [90.7714362827356]
  This paper introduces ERASE, which improves model behavior when new documents are acquired. It incrementally deletes or rewriting other entries in the knowledge base each time a document is added. It improves accuracy relative to conventional retrieval-augmented generation by 7-13% (Mixtral-8x7B) and 6-10% (Llama-3-8B) absolute.
  arXiv  Detail & Related papers  (2024-06-17T17:59:35Z)
• Revisiting Domain-Adaptive 3D Object Detection by Reliable, Diverse and Class-balanced Pseudo-Labeling [38.07637524378327]
  Unsupervised domain adaptation (DA) with the aid of pseudo labeling techniques has emerged as a crucial approach for domain-adaptive 3D object detection. Existing DA methods suffer from a substantial drop in performance when applied to a multi-class training setting. We propose a novel ReDB framework tailored for learning to detect all classes at once.
  arXiv  Detail & Related papers  (2023-07-16T04:34:11Z)
• Composed Image Retrieval with Text Feedback via Multi-grained Uncertainty Regularization [73.04187954213471]
  We introduce a unified learning approach to simultaneously modeling the coarse- and fine-grained retrieval. The proposed method has achieved +4.03%, +3.38%, and +2.40% Recall@50 accuracy over a strong baseline.
  arXiv  Detail & Related papers  (2022-11-14T14:25:40Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.