NTv3 650M pre-training wall-clock time
Training compute of NTv3 650M training run (Boshar et al. 2025)
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.
Descriptive fact transcribed from the pinned source.
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.
6 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Claimed field training_compute Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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 location Section 4.4.4 'Pre-training Compute' (page 34) Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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. |
| Claimed value 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. Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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. |
| Description Descriptive fact transcribed from the pinned source. Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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. |
| Relationship: subject gfm-guide-20261014-config-ntv3-650m-training Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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. |
| Name NTv3 650M pre-training wall-clock time Context-only references | A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 Section 4.4.4 'Pre-training Compute' (page 34) Version: bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: 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. |
Sources and history
Release 2026-10-22-1b7cf027e4f6 · Record review: source checked
1 source record and release history
- A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction (NTv3), bioRxiv v1 · Original source · bioRxiv v1, posted 2025-12-25 (dated 2025-12-22 on the title page)
Technical metadata and extraction receipts
Stable ID: gfm-guide-20261014-claim-ntv3-650m-pretraining-wall-clock
- field
- training_compute
- value
- 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.
- source locator
- Section 4.4.4 'Pre-training Compute' (page 34)
- review
- method: source-hash-verification; ai-assisted-source-review; method note: Each source sentence the claim rests on asserted verbatim in the reviewer's own text extraction (review/check_gfm_guide.py); the claim text was then read against them.; reviewer: claude; reviewer note: Independent Claude (Opus 5.5) review agent for PR #150; did not extract this batch; no human review claimed; date: 2026-10-14; artifact sha256: 374a7177ebf75bf9ce6398624a74e4d588581d54a13ba12263d146b194b961ea; retrieval url: https://www.biorxiv.org/content/biorxiv/early/2025/12/25/2025.12.22.695963.full.pdf; note: Claim matches the source.