---
project: modelmap
document: Phase 1 notes — §4.6 Knowledge localization and editing
author: Simon-Pierre Boucher
contact: contact@spboucher.ai
website: https://modelmap.io
created: 2026-08-12
status: draft
---

# §4.6 — Knowledge localization and editing

## B1–B2. ROME and MEMIT
- **Key papers:** ROME: Meng et al., NeurIPS 2022, https://arxiv.org/abs/2202.05262 (code https://github.com/kmeng01/rome — CUDA-hardcoded in places, no Apple Silicon support); MEMIT: Meng et al., ICLR 2023, https://arxiv.org/abs/2210.07229 (https://github.com/kmeng01/memit)
- **Claims:** causal tracing localizes fact recall to mid-layer MLPs at subject tokens; rank-one (ROME) / spread least-squares (MEMIT, 10k edits) weight updates rewrite facts.
- **Cost:** tracing minutes/fact on Mac at 7B; edits cheap, **but covariance statistics need ~100k Wikipedia forward passes** (precomputed stats exist only for GPT-2/GPT-J; 2025 reduction: https://arxiv.org/pdf/2506.04226).
- **Epistemic status:** Landmark; localization claim debunked-in-part (B8); editing mechanics established.

## B3–B5. PMET, EMMET, AlphaEdit
- **PMET:** Li et al., AAAI 2024, https://arxiv.org/abs/2308.08742 — writes only FFN weights; its finding that MHSA weights need no update is itself interpretability evidence (aligned with B7).
- **EMMET:** Gupta et al., EMNLP Findings 2024, https://arxiv.org/abs/2403.14236 — unifies ROME/MEMIT as one preservation–memorization objective; useful theory for an MLX reimplementation.
- **AlphaEdit:** Fang et al., **ICLR 2025 Outstanding Paper**, https://arxiv.org/abs/2410.02355 (code https://github.com/jianghoucheng/alphaedit) — null-space projection onto preserved-knowledge keys; +36.7% over locate-then-edit baselines; mitigates sequential collapse; one-line projection + one SVD. 2026 reproducibility study broadly confirms but tempers (https://arxiv.org/pdf/2606.26783). Successor: UltraEdit (2025), https://arxiv.org/abs/2505.14679
- **Epistemic status:** Established SOTA line, still fresh.

## B6. Knowledge neurons
- **Key paper:** Dai et al., ACL 2022, https://arxiv.org/abs/2104.08696
- **Critiques:** neurons express token/routing patterns rather than stored knowledge; language-independent and degenerate (redundant) knowledge neurons undermine one-fact-one-neuron (https://arxiv.org/html/2308.13198); attribution pitfalls (EMNLP 2024, https://arxiv.org/abs/2312.12141).
- **Epistemic status:** **Debunked-in-part**; historically important. Mac-feasible to reproduce.

## B7. Storage vs retrieval — the current consensus scaffold
- **Key papers:** Geva et al., EMNLP 2023, https://arxiv.org/abs/2304.14767 (three-stage recall: early-MLP subject enrichment → relation propagation → **attribute extraction by attention heads**; causal via attention knockouts); attribute lens / linear relation decoding: Hernandez et al., ICLR 2024, https://arxiv.org/abs/2308.09124 (~48% of relations well-fit by a single affine map — linearity NOT universal); additive mechanisms: Chughtai, Cooney, Nanda 2024, https://arxiv.org/abs/2402.07321 (recall = sum of qualitatively different mechanisms — complicates clean localization).
- **Consensus:** MLPs store, attention retrieves — established; details (additivity, linearity coverage) contested.
- **Cost:** all Mac-feasible ≤7B.

## B8. The localization-vs-editing critique — **central calibration datum for Q1**
- **Key paper:** Hase et al., NeurIPS 2023, https://arxiv.org/abs/2301.04213
- **Substance:** causal-tracing localization does **not** predict which layer edits best; edit success near-uniform across layers; tracing effects explain a tiny fraction of edit-success variance. **"Edit success" cannot be used as evidence of localization — design Experiment C accordingly.**
- **Related:** *Rebuilding ROME* (collapse partly implementation asymmetry, https://arxiv.org/abs/2403.07175); *Should We Really Edit Language Models?* (evaluation misleading, https://arxiv.org/pdf/2410.18785)
- **Epistemic status:** Established — a causal audit of a causal method. Reproduction Mac-feasible on GPT-2-XL.

## B9. Ripple effects
- **Key paper:** Cohen et al., TACL 2024, https://arxiv.org/abs/2307.12976 — editors fail to propagate logical consequences (multi-hop, sibling, aggregation); **in-context editing beats parametric editors**. 2025–2026 follow-ups: ChainEdit, RippleCOT, https://arxiv.org/pdf/2606.01610
- **Epistemic status:** Established.

## B10. Sequential-editing collapse and general-ability damage
- **Key papers:** Gupta et al. 2024, https://arxiv.org/abs/2401.07453 (gradual forgetting → abrupt collapse); Gu et al., EMNLP 2024, https://arxiv.org/abs/2401.04700 (few edits degrade reasoning/NLI/QA); 2025 energy-regularized line: https://arxiv.org/html/2510.01172v1
- **Consequence:** central caveat for any "editing as causal verification" protocol in modelmap.

## B11. Multilingual knowledge sharing and cross-lingual editing
- **Key papers:** cross-lingual editing: https://arxiv.org/abs/2309.08952 ; language-independent neurons: https://arxiv.org/html/2308.13198 ; MPN: https://arxiv.org/abs/2401.03190 ; language-agnostic factual neurons: https://arxiv.org/abs/2406.16416 ; multilingual recall: https://arxiv.org/abs/2410.14387 ; multilingual pipeline replication (2025): https://arxiv.org/html/2505.20546v2 ; survey (EMNLP 2025): https://arxiv.org/html/2505.14393
- **Substance:** monolingual edits transfer poorly; a shared language-agnostic substrate exists (English-centric hub behavior common); Geva pipeline partially replicates multilingually with language-specific extraction stages. **Directly matches modelmap's FR/EN promptset plans (Experiment B).**
- **Epistemic status:** Sharing established qualitatively; degree and mechanism contested.

## B12. Tooling and Apple Silicon reality check
- **EasyEdit** https://github.com/zjunlp/EasyEdit (ACL 2024) unifies ROME/MEMIT/PMET/AlphaEdit + KnowEdit benchmark; surveys: https://arxiv.org/abs/2310.16218 , https://arxiv.org/pdf/2401.01286
- **All editing repos are CUDA-first; none advertises MPS support.** The math is small-matrix linear algebra — **an MLX/MPS reimplementation is tractable and itself a publishable modelmap artifact ("knowledge editing on consumer Macs")**. The only expensive step is covariance precomputation (hours of forward passes at 7B on 64 GB).

## Implications for modelmap
1. Never infer localization from edit success (Hase). Experiment C's correlational→causal survival rate must use patching/ablation, with editing as a *separate* axis.
2. The storage-vs-retrieval scaffold (B7) gives concrete map types: subject-enrichment maps (MLP), extraction maps (attention heads).
3. Multilingual localization atlas (FR/EN) has solid prior art to build on and clear open questions.
4. MLX knowledge-editing port = concrete tooling contribution.