Configuration

NT-v2-50M-MS

Nucleotide Transformer is a family of DNA encoders pretrained on human or multispecies sequence corpora.

Sources (5)

instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training

2 evaluations · 2 results

How it worksNucleotide Transformer workflow

Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation.

Sources (5)

Overview

Model type

DNA transformer encoder family

Inputs

DNA sequences tokenized into 6-mers, with single-base handling of N and remainder bases.

Outputs

Contextual representations used in specified downstream prediction workflows.

Access

Official project documentation and implementation: https://github.com/instadeepai/nucleotide-transformer

Sources (5)

limited source coverage · Automated source review, 2026-09-16. All specifications and missing details

Evaluations and results

2 evaluations · 2 results. Different protocols are not a single leaderboard.

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: NT-v2-50M-MS	Task: GENEB LINEAR-PROBE: Average macro-MCC across the 13 representative tasks, linear probe Dataset subset: GENEB representative task subset (GENEB split)	0.511 macro_mcc correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source NT-v2-50M-MS on GENEB LINEAR-PROBE: Average macro-MCC across the 13 representative tasks, linear probe Frozen embeddings scored with a probe over the thirteen representative tasks, one from each functional category (Table 7). Aggregation: Not reported GENEB: Why Genomic Models Are Hard to Compare · Table 8, row(NT-v2-50M-MS), column(Linear MCC)
Configuration: NT-v2-50M-MS	Task: GENEB MLP-PROBE: Average macro-MCC across the 13 representative tasks, MLP probe Dataset subset: GENEB representative task subset (GENEB split)	0.521 macro_mcc correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source NT-v2-50M-MS on GENEB MLP-PROBE: Average macro-MCC across the 13 representative tasks, MLP probe Frozen embeddings scored with a probe over the thirteen representative tasks, one from each functional category (Table 7). Aggregation: Not reported GENEB: Why Genomic Models Are Hard to Compare · Table 8, row(NT-v2-50M-MS), column(MLP MCC)

Source checking is not independent reproduction. Release 2026-09-29-06401fd5b220.

Use this model

How it works, versions and access

Related profile: Nucleotide Transformer. This page retains the exact record and its evaluation context.

This configuration

Genomic foundation model evaluated by the GENEB authors with frozen embeddings and a probe.

record: NT-v2-50M-MS
configuration: Not reported
entity type: Configuration

How it works

Nucleotide Transformer is a family of DNA encoders pretrained on human or multispecies sequence corpora. Encoder-only transformers with 6-mer tokens; v1 uses learned positional encodings and v2 uses rotary positions and SwiGLU. The documented inputs are DNA sequences tokenized into 6-mers, with single-base handling of N and remainder bases. The output consists of contextual representations used in specified downstream prediction workflows.

Sources (5)

Versions and reproducibility

NT-v1 human-reference/1000G/multispecies variants and NT-v2 50M/100M/250M/500M. NT-v3 is a separate architecture described elsewhere in the repository. v1: approximately 6kb; v2: 2,048 tokens, approximately 12kb. Exact base count depends on special and ambiguous tokens.

Sources (5)

Strengths, limitations and unresolved questions

Strengths and limitations

Strengths and considerations

The family provides multiple sizes and training corpora, allowing those factors to be compared explicitly.
Sources (5)
instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training

Limitations and conditions

Training corpora and architectures differ across v1 and v2. A family-level name is insufficient to reproduce a score.
Sources (5)
instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training

Profile review details

Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied.

Stable record: discovery-model-nucleotide-transformer

Specifications

Inputs, training, access and other details

Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.

Inputs, outputs and configuration
Property	Description and evidence
Model type	DNA transformer encoder family Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Architecture	Encoder-only transformers with 6-mer tokens; v1 uses learned positional encodings and v2 uses rotary positions and SwiGLU. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Inputs	DNA sequences tokenized into 6-mers, with single-base handling of N and remainder bases. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Outputs	Contextual representations used in specified downstream prediction workflows. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Parameters	50M to 2.5B across the documented v1/v2 family. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Known versions	NT-v1 human-reference/1000G/multispecies variants and NT-v2 50M/100M/250M/500M. NT-v3 is a separate architecture described elsewhere in the repository. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Training data	v1 variants use GRCh38, 3,202 human genomes or 850 multispecies genomes; v2 uses the multispecies corpus. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Training cutoff	The paper specifies human-reference, 1000 Genomes and multispecies training collections by variant. A single latest-deposition date for all sequences is not supplied in the inspected pretraining-data section. · Not reported in inspected sources Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Context limits	v1: approximately 6kb; v2: 2,048 tokens, approximately 12kb. Exact base count depends on special and ambiguous tokens. Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Weights licence	Separate checkpoint-distribution terms are not stated in the inspected release documentation and licence material. The source-code licence alone is not recorded as an explicit weight grant. · Not reported in inspected sources Sources (6) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML; instadeepai/nucleotide-transformer: LICENSE.md · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training; LICENSE.md: licence text
Access	Official project documentation and implementation: https://github.com/instadeepai/nucleotide-transformer Sources (5) instadeepai/nucleotide-transformer: README.md; instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/agro_nucleotide_transformer.md; instadeepai/nucleotide-transformer: docs/segment_nt.md; nt: Journal full-text XML · docs/nucleotide_transformer.md: Model Variants and Sizes, Tokenization and How to use; Nucleotide Transformer paper Methods: Architecture, Pre-training datasets and Training
Code licence	CC-BY-NC-SA-4.0 Sources instadeepai/nucleotide-transformer: LICENSE.md · LICENSE.md: licence text

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.

3 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-29-06401fd5b220
Property and statement	Original source and location	Review and provenance
Relationship: family catalog-model-nt-v2 Individual claims	GENEB: Why Genomic Models Are Hard to Compare Original source ↗ GENEB Table 8 NT-v2-50M-MS; NABench model inventory N.T.v2 50M; official v2-50m-multi-species model card \| Existing reviewed locator: Table 8, row(NT-v2-50M-MS) Version: 2606.04525v1 Retrieved: 2026-09-17T07:56:09.460928+00:00	source checked automated source review · 2026-09-23 Audit details Source review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Same named size/version checkpoint family supported, but exact revision remains unreported. Do not equate weights or adaptations. Field: `links:family:catalog-model-nt-v2` Claim: model-evaluation-identity-1ed4246bb82a6337b56c Source artifact SHA-256: `47975089c0ca738d5e2d6e6ea91cd4e7b80498c3f175803d8ec39ca77aa41629` Hash scope: Exact retrieved primary paper artifact bytes. Inspected artifact
Relationship: family discovery-model-nucleotide-transformer Individual claims	GENEB: Why Genomic Models Are Hard to Compare Original source ↗ mRNABench Table 2/Appendix model inventory; GenomeOcean Table 2; respective model methods; NABench model inventory; GENEB Table 8; source-labelled configuration NT-v2-50M-MS \| Existing reviewed locator: Table 8, row(NT-v2-50M-MS) Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 2606.04525v1 Retrieved: 2026-09-17T07:56:09.460928+00:00	source checked automated source review · 2026-09-23 Audit details Source review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Only family identity is shared. Preserve size, v1/v2, corpus and adaptation distinctions; generic NT summary selects a best-overall family variant, without supplying checkpoint identity. Field: `links:family:discovery-model-nucleotide-transformer` Claim: model-evaluation-identity-44cb3538a9f9dac5a3c1 Source artifact SHA-256: `47975089c0ca738d5e2d6e6ea91cd4e7b80498c3f175803d8ec39ca77aa41629` Hash scope: Exact retrieved primary paper artifact bytes. Inspected artifact
Relationship: family discovery-model-nucleotide-transformer Individual claims	instadeepai/nucleotide-transformer: docs/nucleotide_transformer.md Original source ↗ mRNABench Table 2/Appendix model inventory; GenomeOcean Table 2; respective model methods; NABench model inventory; GENEB Table 8; source-labelled configuration NT-v2-50M-MS \| Existing reviewed locator: Table 8, row(NT-v2-50M-MS) Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 2dc37b86e16a6970fbc731751f7719d9f676f7f9 Retrieved: 2026-09-16T19:46:19.364532+00:00	source checked automated source review · 2026-09-23 Audit details Source review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Only family identity is shared. Preserve size, v1/v2, corpus and adaptation distinctions; generic NT summary selects a best-overall family variant, without supplying checkpoint identity. Field: `links:family:discovery-model-nucleotide-transformer` Claim: model-evaluation-identity-44cb3538a9f9dac5a3c1 Source artifact SHA-256: `ab16d582de98652526b5cebb120eec969328f9db29dc741826bcd81c397e0672` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact

All evidence (CSV)All evidence (JSONL)How to read the evidence

Sources and history

View linked audit checks and correction history

Release 2026-09-29-06401fd5b220 · Record review: source checked

1 source records and release history

GENEB: Why Genomic Models Are Hard to Compare · Original source · 2606.04525v1

Download this release

Technical metadata and extraction receipts

Stable ID: geneb-method-nt-v2-50m-ms

areas: dna-genomes
source locator: Table 8, row(NT-v2-50M-MS)
missing metadata: checkpoint revision: unreported; parameters: unextracted

Related records

Suggest a correction