Pith Number

pith:UUJI3VCD

pith:2023:UUJI3VCD7C2B7TL7EFFQAJ6BIM

not attested not anchored not stored refs pending

A Simple and Effective Pruning Approach for Large Language Models

Anna Bair, J. Zico Kolter, Mingjie Sun, Zhuang Liu

Wanda prunes large language models by removing weights whose magnitudes times input activations are smallest, with no retraining required.

arxiv:2306.11695 v3 · 2023-06-20 · cs.CL · cs.AI · cs.LG

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{UUJI3VCD7C2B7TL7EFFQAJ6BIM}

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

Wanda significantly outperforms the established baseline of magnitude pruning and performs competitively against recent methods involving intensive weight update on LLaMA and LLaMA-2 across various language benchmarks, with no retraining or weight update required.

C2weakest assumption

The central assumption is that the product of weight magnitude and input activation reliably identifies weights whose removal will least affect model performance, motivated by emergent large-magnitude features but without a formal proof that this criterion is optimal or generalizes beyond the tested models.

C3one line summary

Wanda prunes pretrained LLMs by dropping weights with smallest magnitude times activation values per output channel, with no retraining needed and better results than magnitude pruning on language benchmarks.

Formal links

2 machine-checked theorem links

Cited by

42 papers in Pith

Junk DNA Hypothesis: Pruning Small Pre-Trained Weights Irreversibly and Monotonically Impairs "Difficult" Downstream Tasks in LLMs

Condense, Don't Just Prune: Enhancing Efficiency and Performance in MoE Layer Pruning

Will LLMs Scaling Hit the Wall? Breaking Barriers via Distributed Resources on Massive Edge Devices

A3 : an Analytical Low-Rank Approximation Framework for Attention

LLM-Powered AI Agent Systems and Their Applications in Industry

Receipt and verification

First computed	2026-05-17T23:38:47.336614Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

a5128dd443f8b41fcd7f214b0027c1432852a98772cd0d9ff07c4f85d33c8cbd

Aliases

arxiv: 2306.11695 · arxiv_version: 2306.11695v3 · doi: 10.48550/arxiv.2306.11695 · pith_short_12: UUJI3VCD7C2B · pith_short_16: UUJI3VCD7C2B7TL7 · pith_short_8: UUJI3VCD

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/UUJI3VCD7C2B7TL7EFFQAJ6BIM \
  | 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: a5128dd443f8b41fcd7f214b0027c1432852a98772cd0d9ff07c4f85d33c8cbd

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "8bd7eefa2e525be9a8735c1df727f097e7d9a424979146ae0529209eab22af9e",
    "cross_cats_sorted": [
      "cs.AI",
      "cs.LG"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2023-06-20T17:18:20Z",
    "title_canon_sha256": "9b92b75b9a1fafea9f2ef9463f06a2857e2f8610d84409b6122816b94f657316"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2306.11695",
    "kind": "arxiv",
    "version": 3
  }
}