Pith Number

pith:FVQMEHJW

pith:2026:FVQMEHJW7XV5USFM4326TRZSVI

not attested not anchored not stored refs resolved

Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs

Junhyuk Jo, Sai Praneeth Karimireddy, Sunwoo Lee, Vincent-Daniel Yun

A closed-form linear operator derived from calibration data can reconstruct the hidden-state mismatch caused by removing entire layers from large language models.

arxiv:2605.15491 v1 · 2026-05-15 · cs.LG · cs.AI · cs.PF

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{FVQMEHJW7XV5USFM4326TRZSVI}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

Our method derives a closed-form optimal linear operator from a small calibration set to reconstruct the activation discrepancy introduced by the pruned layers. We show that this solution corresponds to the unconstrained optimum of the alignment objective, whereas existing methods are restricted to constrained solutions over limited operator subspaces.

C2weakest assumption

The activation discrepancy introduced by pruned layers can be effectively reconstructed by a single linear operator fitted on a small calibration set, and that this operator remains effective across the full range of inputs the model will see at inference time (abstract, paragraph on boundary activation alignment problem).

C3one line summary

Ghosted Layers recovers accuracy in layer-pruned LLMs via a closed-form unconstrained linear operator that aligns boundary activations using a small calibration set.

References

39 extracted · 39 resolved · 7 Pith anchors

[1] Fluctuation-based adaptive structured pruning for large language models 2024

[2] Croci, Marcelo Gennari do Nascimento, Torsten Hoefler, and James Hensman 2024

[3] Language models are few-shot learners 1901

[4] Streamlining redundant lay- ers to compress large language models 2025

[5] A simple linear patch revives layer-pruned large language models 2026

Receipt and verification

First computed	2026-05-20T00:01:01.400278Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

2d60c21d36fdebda48ace6f5e9c732aa26c40b36f80c02fa94f26368567768d6

Aliases

arxiv: 2605.15491 · arxiv_version: 2605.15491v1 · doi: 10.48550/arxiv.2605.15491 · pith_short_12: FVQMEHJW7XV5 · pith_short_16: FVQMEHJW7XV5USFM · pith_short_8: FVQMEHJW

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/FVQMEHJW7XV5USFM4326TRZSVI \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 2d60c21d36fdebda48ace6f5e9c732aa26c40b36f80c02fa94f26368567768d6

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "00d5746c49ec258ce89e12617597256c816036e83c929ff201d3bd8904833f6c",
    "cross_cats_sorted": [
      "cs.AI",
      "cs.PF"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-15T00:15:16Z",
    "title_canon_sha256": "9c8cd4029604923b37bb168feb27034ea5bcf63ae83c22f39c68753283296df3"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.15491",
    "kind": "arxiv",
    "version": 1
  }
}