Relevance of Fine-tuning wall time to convergence on the Enformer eQTL dataset (Hegde et al. 2025 Methods) to "Select an execution workflow for large-scale diagnostic genomics"
- Use case
- Select an execution workflow for large-scale diagnostic genomics
- Assessed by
- Fine-tuning wall time to convergence on the Enformer eQTL dataset (Hegde et al. 2025 Methods)
- Relevance
- Proxy evidence
- Endpoint
- Approximate wall time to fine-tune with LoRA (r=1) to validation-loss convergence on the eQTL training split: Nucleotide Transformer 500M human-ref on 2 × NVIDIA TITAN Xp and pruned DNABERT-2 on 2 × Quadro RTX 8000
Proxy. Approximate wall time to fine-tune two genomic foundation models to convergence on a human variant task, which informs the one-off adaptation cost before a model can run in a workflow. It is not a per-sample execution measurement, and the two values come from different GPUs, so they compare machines as much as models.
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.
18 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
Citation locations1 values
| Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Comparison group hegde2025-eqtl-convergence Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Comparison title Fine-tuning time to convergence on the Enformer eQTL task (Hegde et al. 2025) Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
Constraints3 values
| Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Endpoint Approximate wall time to fine-tune with LoRA (r=1) to validation-loss convergence on the eQTL training split: Nucleotide Transformer 500M human-ref on 2 × NVIDIA TITAN Xp and pruned DNABERT-2 on 2 × Quadro RTX 8000 Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | 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 field links:assessed_by:gfm-guide-20261014-protocol-hegde-eqtl-finetune-convergence-time Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Headline metric runtime Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
Limitations6 values
| Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
Pinned records7 values
| Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Rationale Proxy. Approximate wall time to fine-tune two genomic foundation models to convergence on a human variant task, which informs the one-off adaptation cost before a model can run in a workflow. It is not a per-sample execution measurement, and the two values come from different GPUs, so they compare machines as much as models. Context-only references | Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3 Version: Genes 16(11):1358, 2025; PMC12652071 full-text XML | 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
- Reaping the Fruits of LLM Pruning: Towards Small Language Models for Efficient Non-Coding Variant Effect Prediction · Original source · Genes 16(11):1358, 2025; PMC12652071 full-text XML
Technical metadata and extraction receipts
Stable ID: use-case-mapping-gfm-guide-20261014-hegde2025-eqtl-convergence
- field
- links:assessed_by:gfm-guide-20261014-protocol-hegde-eqtl-finetune-convergence-time
- value
- gfm-guide-20261014-protocol-hegde-eqtl-finetune-convergence-time
- relevance
- proxy
- endpoint
- Approximate wall time to fine-tune with LoRA (r=1) to validation-loss convergence on the eQTL training split: Nucleotide Transformer 500M human-ref on 2 × NVIDIA TITAN Xp and pruned DNABERT-2 on 2 × Quadro RTX 8000
- rationale
- Proxy. Approximate wall time to fine-tune two genomic foundation models to convergence on a human variant task, which informs the one-off adaptation cost before a model can run in a workflow. It is not a per-sample execution measurement, and the two values come from different GPUs, so they compare machines as much as models.
- constraints
- Inspect every linked evaluation's source locator before citing a result.; Do not combine this mapping's evaluations with any other protocol's results; hardware, inputs, software versions and timing boundaries differ between sources.; Do not compare the two rows as a model comparison; they ran on different GPUs.
- limitations
- Approximate prose values ('approximately 19 h'; 'around 1 h per pruned model on average'), not tabulated per run.; Different GPUs for the two models (2 × NVIDIA TITAN Xp; 2 × Quadro RTX 8000).; The DNABERT-2 value is an average over pruned models with one encoder layer removed, not the full model.; The models differ in size (DNABERT-2 117 million and the Nucleotide Transformer 480 million parameters, Table 2).; Fine-tuning is LoRA through peft with r=1 (Appendices A.2 and A.3), not full fine-tuning.; The authors state that the Nucleotide Transformer could not be trained to convergence three times per version, so its value rests on fewer runs; the number is not stated.
- revision
- 1
- reason
- Add relevance judgements for the execution and compute protocols of the genomic foundation model guide sources batch (data/omics/gfm-guide-sources-20261014, issue #162), independently reviewed in PR #169.
- citation locators
- source id: gfm-guide-20261014-source-hegde2025; locator: Methods paragraph 3; Table 2; Appendices A.2 and A.3
- source locator
- gfm-guide-20261014-source-hegde2025: Methods paragraph 3; Table 2; Appendices A.2 and A.3
- comparison group
- hegde2025-eqtl-convergence
- comparison title
- Fine-tuning time to convergence on the Enformer eQTL task (Hegde et al. 2025)
- headline metric
- runtime
- reviewed evaluations
- gfm-guide-20261014-eval-hegde-dnabert-2-pruned-convergence; gfm-guide-20261014-eval-hegde-nt-500m-human-ref-convergence
- review
- method: source-hash-verification; independent-cell-check; scope-review; reviewer: claude; reviewer note: Claude (Opus 5.5) review agent review-169, independent of the agent that drafted these judgements for issue #162; no human review claimed; reviewed at: 2026-10-14T19:00:00Z; note: Independent relevance review: grade, endpoint, rationale, constraints, limitations and locators checked against the use case and the source text; every value of the listed evaluations checked against the source. No execution, reproduction, qualified human scientific review or clinical validation. See docs/reviews/use-cases/diagnostic-genomics-model-execution-judgements-2026-10-14.md. Reviewer added LoRA and model size to the endpoint and limitations.
- pins
- record id: gfm-guide-20261014-eval-hegde-dnabert-2-pruned-convergence; fields: status; links; attributes.origin; attributes.comparison; sha256: deb5669cfaeb3573f4a37b1af7be7bf4c52268aff59caa1d4211ed7baab2b3cd; record id: gfm-guide-20261014-eval-hegde-nt-500m-human-ref-convergence; fields: status; links; attributes.origin; attributes.comparison; sha256: 560617de3dd0d9aab4d3a8e9cefb18b950e2746f0d34a7e5ab182d9763ad15f9; record id: gfm-guide-20261014-protocol-hegde-eqtl-finetune-convergence-time; fields: status; name; description; facets; source_ids; links; attributes; sha256: 0e64ea2c8406ef1d420d27a3177a142adf8483240553fc91c461cc293c716bf4; record id: gfm-guide-20261014-result-hegde-dnabert-2-pruned-convergence-time; fields: status; attributes.metric; attributes.metric_qualifier; attributes.printed_value; attributes.numeric_value; sha256: 308c42f81a26b58300230cc1a52d060681ef999c0872c2c9407968ec7dd90282; record id: gfm-guide-20261014-result-hegde-nt-convergence-time; fields: status; attributes.metric; attributes.metric_qualifier; attributes.printed_value; attributes.numeric_value; sha256: 518cf68ce2568917b9a466f16db19b0a90cf1196123f1eddb78432842284ddd3; record id: gfm-guide-20261014-source-hegde2025; fields: status; attributes.artifact_sha256; sha256: 4ec6ee59ca3c282cc9f37731f592837db20813b4ad179f1960acfe6782221a4e; record id: use-case-diagnostic-genomics-model-execution; fields: status; name; description; facets; attributes.question; attributes.decision; attributes.inputs; attributes.output; attributes.exclusions; attributes.setting; attributes.clinical_scope; attributes.intended_users; sha256: 1c2a12431556cc6581f5aab019cf91576d2393387b88a8a2bea35ea2a3ee4a70