Pith Number

pith:MGKPCIXJ

pith:2023:MGKPCIXJYYEXICVREX4KUQJQBZ

not attested not anchored not stored refs resolved

GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints

Federico Lebr\'on, James Lee-Thorp, Joshua Ainslie, Michiel de Jong, Sumit Sanghai, Yury Zemlyanskiy

Uptraining multi-head attention checkpoints to grouped-query attention recovers near-original quality with only 5% additional compute and achieves multi-query inference speeds.

arxiv:2305.13245 v3 · 2023-05-22 · cs.CL · 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{MGKPCIXJYYEXICVREX4KUQJQBZ}

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

We show that uptrained GQA achieves quality close to multi-head attention with comparable speed to MQA.

C2weakest assumption

The uptraining recipe with 5% compute is sufficient to recover near-original quality without task-specific degradation or architecture-dependent failures.

C3one line summary

Uptraining multi-head transformer checkpoints to grouped-query attention models achieves near multi-head quality at multi-query inference speeds using 5% additional compute.

References

40 extracted · 40 resolved · 11 Pith anchors

[1] James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

[2] Jonathan Heek and Anselm Levskaya and Avital Oliver and Marvin Ritter and Bertrand Rondepierre and Andreas Steiner and Marc van

[3] Alexey Romanov and Chaitanya Shivade

[4] Kingma and Jimmy Ba , editor = 2015

[5] Adafactor: Adaptive Learning Rates with Sublinear Memory Cost , booktitle = 2018

Formal links

2 machine-checked theorem links

Cited by

136 papers in Pith

Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

Lifelong In-Context Learning with Transformers Requires Parametric Forms of Attention

ZONOS2 Technical Report

Scaling Laws for Task-Specific LLM Distillation

Grouped Query Experts: Mixture-of-Experts on GQA Self-Attention

Receipt and verification

First computed	2026-07-05T07:27:25.130641Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b

Aliases

arxiv: 2305.13245 · arxiv_version: 2305.13245v3 · doi: 10.48550/arxiv.2305.13245 · pith_short_12: MGKPCIXJYYEX · pith_short_16: MGKPCIXJYYEXICVR · pith_short_8: MGKPCIXJ

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/MGKPCIXJYYEXICVREX4KUQJQBZ \
  | 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: 6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "84953e0a13173d207eb3658876e04a86f84bea9bbd978969ec905a88c7e6e302",
    "cross_cats_sorted": [
      "cs.LG"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2023-05-22T17:16:38Z",
    "title_canon_sha256": "7929c6e9d522dca9ceba7ad6e7fc9a8b3efad9d631da45c97403d77cf800d192"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2305.13245",
    "kind": "arxiv",
    "version": 3
  }
}