NT 2.5B, training with FlashAttention and checkpointing (Saadat and Fellay 2026)
Training of NT 2.5B with FlashAttention and layer-wise activation checkpointing, without ZeRO offload, at sequence length 1000 and batch size 16.
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.
Evaluations and results
1 evaluation · 1 result. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: NT 2.5B, training with FlashAttention and checkpointing (Saadat and Fellay 2026) | Protocol: Training peak GPU memory under progressive optimizations, sequence length 1000, batch size 16 (Saadat and Fellay 2026 Figure 3) Dataset: Synthetic uniform random DNA sequences, seed 42 (Saadat and Fellay 2026) | 33.9 Peak memory (Training, sequence length 1000, batch size 16, stage + Checkpointing (FlashAttention and activation checkpointing)) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceNT 2.5B training peak memory, stage + Checkpointing saadat-fellay-20261015-protocol-training-peak-memory Aggregation: Not reported Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) · Section 3.3 'ZeRO offload' paragraph (page 6): 'the NT 2.5B checkpoint (33.88 → 11.49 GB)'; stage + Checkpointing (FlashAttention and activation checkpointing) of Figure 3 |
Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.
Use this configuration
How it works, versions and access
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
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
Release 2026-10-22-1b7cf027e4f6 · Record review: source checked
1 source record and release history
- Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) · Original source · OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09)
Technical metadata and extraction receipts
Stable ID: saadat-fellay-20261015-config-nt-2-5b-training-flashattention-checkpointing
- areas
- dna-genomes
- contexts
- research
- method types
- foundation_model
- reported name
- NT 2.5B
- model family
- Nucleotide Transformer
- hardware
- description: NVIDIA H100 (section 2.4); 'H100 94GB GPUs' (section 4)
- source locator
- Section 2.2 'Models'; Section 2.3; Section 2.4
- provenance note
- Implementation and measurements by Saadat and Fellay (EPFL), who did not develop the model. Checkpoint identity is not stated, so the configuration stays at family level.
- missing metadata
- checkpoint: reason: unreported; note: The paper names the family and parameter count only. No Hugging Face repository, revision or pretraining dataset is given (several GENA-LM and Nucleotide Transformer checkpoints share these sizes).; version: reason: unreported; note: No code release or software versions are given in the PDF
- fine tuning
- description: FlashAttention and layer-wise activation checkpointing, without ZeRO offload; whether LoRA (the paper's fine-tuning default: rank 16 on query, key and value) was active in this ablation is not stated
- training setting
- Figure 3 stage '+ Checkpointing', cumulative over the previous stages
- review
- reviewer: claude; reviewer note: Claude review agent; no human review claimed. Independent Claude (Opus 5.5) review agent, separate from the extraction agent that built this batch. Automated review only. Transcription, not reproduction.; reviewed at: 2026-10-15T22:00:00Z; artifact sha256: 4bc2452110f280a7ba8baa4c7c28beac89769511019666bc77fc33af9b3bcd60; retrieval url: https://openreview.net/pdf?id=NUleWh1GFx; method: source-hash-verification; pdf-text-parse; ai-assisted-source-review; scope-review; method note: Descriptive fields, identity links and missing-metadata reasons checked against the PDF.; note: Checked; see the dated review.
Related records
- configuration of: Nucleotide Transformer
- system: NT 2.5B training peak memory, stage + Checkpointing