Protocol

Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2)

Evidence concern: excluded from comparisons

Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer'

GPUs, wall-clock time and GPU-hours of pretraining for DNABERT, Nucleotide Transformer 2.5B and two small HyenaDNA models.

4 evaluations · 8 results

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

4 recorded evaluations, 8 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

4 evaluations · 8 results. Different protocols are not a single leaderboard.

Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) · Dataset: Pretraining corpora of the models in HyenaDNA Table A.2

Sorted by Accelerator time (Pretraining, total) (lower is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	Accelerator time (Pretraining, total)	Runtime (Pretraining, wall clock)
HyenaDNA 1.6M pretraining (HyenaDNA Table A.2)Evaluation	1.3 (best in column)	80
HyenaDNA 436K pretraining (HyenaDNA Table A.2)Evaluation	1.3 (best in column)	80
DNABERT 110M pretraining (HyenaDNA Table A.2)Evaluation	12,000	25 (best in column)
Nucleotide Transformer 2.5B pretraining (HyenaDNA Table A.2)Evaluation	215,000	28

All 8 result rows with coverage, uncertainty and sources

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: DNABERT 110M pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	12,000 Accelerator time (Pretraining, total) hour · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source DNABERT 110M pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'GPU-hrs', column 1 of 4 ('DNABERT', 110M)
Configuration: DNABERT 110M pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	25 Runtime (Pretraining, wall clock) day · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source DNABERT 110M pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'Wall clock', column 1 of 4 ('DNABERT', 110M)
Configuration: HyenaDNA 1.6M pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	1.3 Accelerator time (Pretraining, total) hour · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source HyenaDNA 1.6M pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'GPU-hrs', column 4 of 4 ('HyenaDNA', 1.6M)
Configuration: HyenaDNA 1.6M pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	80 Runtime (Pretraining, wall clock) minute · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source HyenaDNA 1.6M pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'Wall clock', column 4 of 4 ('HyenaDNA', 1.6M)
Configuration: HyenaDNA 436K pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	1.3 Accelerator time (Pretraining, total) hour · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source HyenaDNA 436K pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'GPU-hrs', column 3 of 4 ('HyenaDNA', 436K)
Configuration: HyenaDNA 436K pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	80 Runtime (Pretraining, wall clock) minute · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source HyenaDNA 436K pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'Wall clock', column 3 of 4 ('HyenaDNA', 436K)
Configuration: Nucleotide Transformer 2.5B pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	215,000 Accelerator time (Pretraining, total) hour · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source Nucleotide Transformer 2.5B pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'GPU-hrs', column 2 of 4 ('Nucleotide Transformer', 2.5B)
Configuration: Nucleotide Transformer 2.5B pretraining (HyenaDNA Table A.2)	Protocol: Pretraining GPUs and runtime for short-range models (HyenaDNA Table A.2) Dataset: Pretraining corpora of the models in HyenaDNA Table A.2	28 Runtime (Pretraining, wall clock) day · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table A.2 GPU-hours for the DNABERT and Nucleotide Transformer columns do not equal the GPUs × wall clock printed in the same table: 8 GPUs × 25 days is 4,800 GPU-hours against 12,000 printed, and 128 GPUs × 28 days is 86,016 GPU-hours against 215,000 printed, both about 2.5 times higher. The HyenaDNA columns agree (1 GPU × 80 min, about 1.3 GPU-hours). The Nucleotide Transformer article (Methods 'Hardware') also gives 128 A100 GPUs for 28 days. The table cites no source for these columns. Values are kept as printed and should not be compared until the discrepancy is resolved.HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), rows 'GPUs', 'Wall clock' and 'GPU-hrs', columns 'DNABERT' and 'Nucleotide Transformer' Methods, coverage and source Nucleotide Transformer 2.5B pretraining compute (HyenaDNA Table A.2) gfm-guide-20261014-protocol-hyenadna-pretraining-compute Aggregation: Not reported HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Table A.2 (page 17), row 'Wall clock', column 2 of 4 ('Nucleotide Transformer', 2.5B)

Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.

Methods and evaluation design

Procedure, tasks and evaluated configurations

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Baseline coverage

Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.

0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.

No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.

Author-reported evaluations: 2
paper compilation: 2

Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.

Null control

Proposed control: requires review

Select a task-valid null control after reviewing inputs and metric

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Conventional reference

Proposed control: requires review

Select an upstream conventional reference after reviewing the full protocol

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)

Coverage is derived from release 2026-10-22-1b7cf027e4f6. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run instructions

No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.

Strengths, limitations and unresolved questions

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

0 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-10-22-1b7cf027e4f6
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-22-1b7cf027e4f6 · Record review: source checked

1 source record and release history

HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution, arXiv v2 · Original source · arXiv 2306.15794v2

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: gfm-guide-20261014-protocol-hyenadna-pretraining-compute

areas: dna-genomes
contexts: research
protocol: Pretraining cost as printed: GPUs used, wall-clock time and GPU-hours per model. HyenaDNA rows are the authors' own runs; the DNABERT and Nucleotide Transformer rows are stated without a cited source in the table.
version: arXiv v2 Table A.2
metric: accelerator-time
limitations: Mixes GPU types (RTX 2080 Ti, A100 80GB, A100 40GB), so GPU-hours are not comparable in cost.; The two HyenaDNA models are tiny (436K and 1.6M parameters) next to 110M and 2.5B, and pretrain on different data and sequence lengths.
source locator: Table A.2 (page 17)

Related records

uses data: Pretraining corpora of the models in HyenaDNA Table A.2
assessment: DNABERT 110M pretraining compute (HyenaDNA Table A.2)
assessment: HyenaDNA 1.6M pretraining compute (HyenaDNA Table A.2)
assessment: HyenaDNA 436K pretraining compute (HyenaDNA Table A.2)
assessment: Nucleotide Transformer 2.5B pretraining compute (HyenaDNA Table A.2)
assessed by: Select an execution workflow for large-scale diagnostic genomics

Suggest a correction