reported_finding: Saadat and Fellay 2026
Reported finding of Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop)
Under 8-bit and 4-bit quantization, embedding cosine similarity to BF16 stays above 0.98 for GENA-LM 110M, GENA-LM 336M and NT 500M; for NT 2.5B quantization lowers it to roughly 0.9.
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.
7 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Claimed field reported_finding Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
Limitations2 values
| Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Source location Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Claimed value Under 8-bit and 4-bit quantization, embedding cosine similarity to BF16 stays above 0.98 for GENA-LM 110M, GENA-LM 336M and NT 500M; for NT 2.5B quantization lowers it to roughly 0.9. Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Description Descriptive fact transcribed from the pinned source. Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Relationship: subject saadat-fellay-20261015-source-paper Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Name reported_finding: Saadat and Fellay 2026 Context-only references | Optimizing genomic language models for efficient training, fine-tuning, and inference (Saadat and Fellay, ICLR 2026 workshop) Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.' Version: OpenReview PDF headed 'Published as a workshop paper at ICLR 2026' (11 pages; PDF CreationDate 2026-03-09) | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
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-claim-quantization-embedding-similarity
- field
- reported_finding
- value
- Under 8-bit and 4-bit quantization, embedding cosine similarity to BF16 stays above 0.98 for GENA-LM 110M, GENA-LM 336M and NT 500M; for NT 2.5B quantization lowers it to roughly 0.9.
- source locator
- Section 3.5 'Fidelity' paragraph (page 6): 'All models except NT 2.5B maintain embedding similarity above 0.98. NT 2.5B is more sensitive: quantization lowers similarity to roughly 0.9.'
- 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; rendered-source-visual-check; ai-assisted-source-review; method note: Source sentences asserted in the pdftotext output by review/review.py; Figures 2-5 viewed to check the claim's scope.; note: Correct; Figure 5 left panel is consistent with the bounds.
- limitations
- Bounds and an approximate value, not per-model results; Figure 5 left panel plots the bars without values.; The paper does not say which of 8-bit and 4-bit gives 'roughly 0.9' for NT 2.5B.