Source

A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1

bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) · Preprint

DOI: 10.64898/2025.12.22.695963

What this source contributed

Records in this release that cite this source. A record can cite several sources.

3 configurations
1 protocol
1 dataset
3 evaluations
13 results
3 evidence claims

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Value	Metric	Model or configuration	Dataset	Evaluation
40 hour	Accelerator time (Training stage 'Length ext. 1-4' (200B tokens))	NTv3 8M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 8M training GPU-hours (Boshar et al. 2025 Table 5)
40 hour	Accelerator time (Training stage 'Length ext. 5-7' (150B tokens))	NTv3 8M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 8M training GPU-hours (Boshar et al. 2025 Table 5)
1,608 hour	Accelerator time (Training stage 'Pre-training 8 kb' (10.5T tokens))	NTv3 8M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 8M training GPU-hours (Boshar et al. 2025 Table 5)
40 hour	Accelerator time (Training stage 'Length ext. 1-4' (200B tokens))	NTv3 100M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 100M training GPU-hours (Boshar et al. 2025 Table 5)
48 hour	Accelerator time (Training stage 'Length ext. 5-7' (150B tokens))	NTv3 100M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 100M training GPU-hours (Boshar et al. 2025 Table 5)
1,536 hour	Accelerator time (Training stage 'Post-training 1' (1.2T tokens))	NTv3 100M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 100M training GPU-hours (Boshar et al. 2025 Table 5)
192 hour	Accelerator time (Training stage 'Post-training 2' (54B tokens))	NTv3 100M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 100M training GPU-hours (Boshar et al. 2025 Table 5)
3,000 hour	Accelerator time (Training stage 'Pre-training 8 kb' (10.5T tokens))	NTv3 100M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 100M training GPU-hours (Boshar et al. 2025 Table 5)
88 hour	Accelerator time (Training stage 'Length ext. 1-4' (200B tokens))	NTv3 650M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 650M training GPU-hours (Boshar et al. 2025 Table 5)
80 hour	Accelerator time (Training stage 'Length ext. 5-7' (150B tokens))	NTv3 650M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 650M training GPU-hours (Boshar et al. 2025 Table 5)
1,824 hour	Accelerator time (Training stage 'Post-training 1' (1.2T tokens))	NTv3 650M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 650M training GPU-hours (Boshar et al. 2025 Table 5)
192 hour	Accelerator time (Training stage 'Post-training 2' (54B tokens))	NTv3 650M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 650M training GPU-hours (Boshar et al. 2025 Table 5)
6,736 hour	Accelerator time (Training stage 'Pre-training 8 kb' (10.5T tokens))	NTv3 650M training run (Boshar et al. 2025)	NTv3 training data, all stages (Boshar et al. 2025)	NTv3 650M training GPU-hours (Boshar et al. 2025 Table 5)

Claims

Statements this source supports, each with where in the source it was checked.

NTv3 650M pre-training wall-clock time: Training compute Pre-training NTv3 650M on the full 10.5 trillion tokens took over nine months of H100 GPU-hours; on 4 nodes (32 GPUs) the wall-clock time was just under nine days. Throughput was about 3.5M tokens/s for 650M, 7.8M for 100M and 14.5M for 8M. (Section 4.4.4 'Pre-training Compute' (page 34))
NTv3 inference hardware: Hardware requirement The authors state that even the largest NTv3 model is designed to run inference for full megabase windows on a single A100 or H100 GPU, and that NTv3 8M can run inference for sequences of several tens of kilobases on a laptop CPU within seconds in mixed precision. (Results paragraph spanning pages 4 and 6 (page 5 is a figure))
Relevance of NTv3 H100 GPU-hours per training stage (Boshar et al. 2025 Table 5) to "Select an execution workflow for large-scale diagnostic genomics": Assessed by NTv3 H100 GPU-hours per training stage (Boshar et al. 2025 Table 5) (gfm-guide-20261014-source-ntv3-v1: Table 5 and section 4.4.4 'Training Infrastructure' (page 34))

Models, benchmarks and other records

Configurations (3)

Protocols (1)

NTv3 H100 GPU-hours per training stage (Boshar et al. 2025 Table 5)

Datasets (1)

NTv3 training data, all stages (Boshar et al. 2025)

Evaluations (3)

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.

19 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
attributes.artifact_sha256 374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.artifact_sha256` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Artifact url https://www.biorxiv.org/content/biorxiv/early/2025/12/25/2025.12.22.695963.full.pdf Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.artifact_url` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Doi 10.64898/2025.12.22.695963 Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.doi` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Hash scope SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.hash_scope` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Licence CC-BY-NC-ND-4.0 Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.licence` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Media type application/pdf Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.media_type` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Publication status preprint Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.publication_status` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Retrieval note First request returned HTTP 429; the retry two minutes later returned the PDF (107 pages). Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.retrieval_note` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Retrieved at 2026-10-14T12:05:27Z Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.retrieved_at` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact
Url https://doi.org/10.64898/2025.12.22.695963 Source metadata	A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Original source ↗ No field-specific location recorded Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) Retrieved: 2026-10-14T12:05:27Z	catalogued No individual claim review recorded Audit details Field: `attributes.url` Source artifact SHA-256: `374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy. Inspected artifact

How to read the evidence

Sources and history

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

Preprint PDF retrieved and hashed for the genomic foundation model guide sources pass (issue #138).

Release history

No supporting source is linked yet.

Read original source

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: gfm-guide-20261014-source-ntv3-v1

areas: dna-genomes
contexts: research
artifact url: https://www.biorxiv.org/content/biorxiv/early/2025/12/25/2025.12.22.695963.full.pdf
retrieved at: 2026-10-14T12:05:27Z
artifact sha256: 374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea
url: https://doi.org/10.64898/2025.12.22.695963
doi: 10.64898/2025.12.22.695963
publication status: preprint
licence: CC-BY-NC-ND-4.0
version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page)
media type: application/pdf
retrieval note: First request returned HTTP 429; the retry two minutes later returned the PDF (107 pages).
hash scope: SHA-256 of one retrieval of the bioRxiv PDF (2026-10-14T12:05:27Z). bioRxiv stamps the PDF on each download (XMP ModifyDate, MetadataDate, DocumentID and InstanceID, the Info ModDate and the trailer /ID), so later retrievals differ. The extractor's copy was not retained, so the pinned bytes could not be re-verified. The independent reviewer's retrieval on 2026-10-14 (15,535,038 bytes, 107 pages, ModDate D:20261014041002-07'00') had SHA-256 082ea9b096d634cad3fc568465ef887d2529060909777685261cfc0bb43fcbee; all 13 Table 5 cells and the quoted section 4.4.4 and Results sentences match in that copy.

Related records

Suggest a correction