Protocol

GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3)

Embedding extraction on one H100 at three model scales with BF16 and FP8 weights.

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.

6 recorded evaluations, 21 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

6 evaluations · 21 results. Different protocols are not a single leaderboard.

Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) · Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)

Sorted by Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels)	Inference throughput (Embedding extraction; input tokens per second)	Peak memory (Embedding extraction; peak GPU memory allocated)	Throughput change (Embedding extraction; relative to the BF16 baseline of the same model scale)
GenomeOcean-4B, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)Evaluation	0.5280 ± 0.0653 (best in column)	22,401 ± 204	67.9	4.7% (best in column)
GenomeOcean-4B, BF16, PyTorch embedding extraction (Yu et al. 2026)Evaluation	0.5278 ± 0.0672	21,386 ± 126	72.1	Not reported
GenomeOcean-100M, BF16, PyTorch embedding extraction (Yu et al. 2026)Evaluation	0.3608 ± 0.0619	613,371 ± 22,439 (best in column)	21.1	Not reported
GenomeOcean-100M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)Evaluation	0.3533 ± 0.0588	460,733 ± 1,694	21 (best in column)	−24.9%
GenomeOcean-500M, BF16, PyTorch embedding extraction (Yu et al. 2026)Evaluation	0.3345 ± 0.0479	230,523 ± 651	47.2	Not reported
GenomeOcean-500M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)Evaluation	0.3273 ± 0.0508	204,973 ± 777	46.7	−11.1%

All 21 result rows with coverage, uncertainty and sources

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: GenomeOcean-100M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.3608 ± 0.0619 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0619 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'BF16', column 'DBSCAN ARI'
Configuration: GenomeOcean-100M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	21.1 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'BF16', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-100M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	613,371 ± 22,439 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 22439 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'BF16', column 'Throughput (tok/s)'
Configuration: GenomeOcean-100M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.3533 ± 0.0588 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0588 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'FP8 (E4M3)', column 'DBSCAN ARI'
Configuration: GenomeOcean-100M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	21 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'FP8 (E4M3)', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-100M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	460,733 ± 1,694 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 1694 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'FP8 (E4M3)', column 'Throughput (tok/s)'
Configuration: GenomeOcean-100M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	−24.9% Throughput change (Embedding extraction; relative to the BF16 baseline of the same model scale) percent · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-100M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '100M', Precision 'FP8 (E4M3)', column 'Change'
Configuration: GenomeOcean-4B, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.5278 ± 0.0672 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0672 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'BF16', column 'DBSCAN ARI'
Configuration: GenomeOcean-4B, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	72.1 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'BF16', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-4B, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	21,386 ± 126 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 126 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'BF16', column 'Throughput (tok/s)'
Configuration: GenomeOcean-4B, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.5280 ± 0.0653 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0653 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'FP8 (E4M3)', column 'DBSCAN ARI'
Configuration: GenomeOcean-4B, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	67.9 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'FP8 (E4M3)', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-4B, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	22,401 ± 204 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 204 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'FP8 (E4M3)', column 'Throughput (tok/s)'
Configuration: GenomeOcean-4B, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	4.7% Throughput change (Embedding extraction; relative to the BF16 baseline of the same model scale) percent · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-4B FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '4B', Precision 'FP8 (E4M3)', column 'Change'
Configuration: GenomeOcean-500M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.3345 ± 0.0479 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0479 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'BF16', column 'DBSCAN ARI'
Configuration: GenomeOcean-500M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	47.2 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'BF16', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-500M, BF16, PyTorch embedding extraction (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	230,523 ± 651 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 651 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M BF16 embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'BF16', column 'Throughput (tok/s)'
Configuration: GenomeOcean-500M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	0.3273 ± 0.0508 Adjusted rand index (DBSCAN clustering of embeddings against GTDB species labels) unitless · higher Uncertainty: SD 0.0508 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'FP8 (E4M3)', column 'DBSCAN ARI'
Configuration: GenomeOcean-500M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	46.7 Peak memory (Embedding extraction; peak GPU memory allocated) gigabyte · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'FP8 (E4M3)', column 'Peak VRAM (GB)'
Configuration: GenomeOcean-500M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	204,973 ± 777 Inference throughput (Embedding extraction; input tokens per second) sample-per-second · higher Uncertainty: SD 777 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'FP8 (E4M3)', column 'Throughput (tok/s)'
Configuration: GenomeOcean-500M, FP8 (E4M3) static weights, PyTorch with FBGEMM (Yu et al. 2026)	Protocol: GenomeOcean embedding extraction throughput, peak VRAM and binning fidelity, BF16 versus FP8 weights (Yu et al. 2026 Table 3) Dataset: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)	−11.1% Throughput change (Embedding extraction; relative to the BF16 baseline of the same model scale) percent · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source GenomeOcean-500M FP8 (E4M3) embedding extraction (Yu et al. 2026) model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8 Aggregation: Not reported FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Table 3, row Model '500M', Precision 'FP8 (E4M3)', column 'Change'

Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.

Methods and evaluation design

Procedure, tasks and evaluated configurations

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Baseline coverage

Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.

0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.

No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.

Author-reported evaluations: 6

Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.

Null control

Proposed control: requires review

Select a task-valid null control after reviewing inputs and metric

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Conventional reference

Proposed control: requires review

Select an upstream conventional reference after reviewing the full protocol

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)

Coverage is derived from release 2026-10-22-1b7cf027e4f6. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run instructions

No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.

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

Claims, original sources and review scope · Release 2026-10-22-1b7cf027e4f6
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-22-1b7cf027e4f6 · Record review: source checked

1 source record and release history

FP8 Inference in Genomic Foundation Models: Theoretical vs. Realized Speedups on GenomeOcean · Original source · bioRxiv preprint version 1, posted 2026-08-14 (PDF footer); bioRxiv API date 2026-08-18

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: model-execution-20261011-protocol-yu2026-genomeocean-embedding-fp8

areas: dna-genomes
contexts: research
protocol: Single causal forward pass over batched tokenized fragments; final-layer states mean-pooled. Throughput in input tokens per second at a fixed batch size per scale; peak memory from torch.cuda.max_memory_allocated(); embeddings clustered with DBSCAN (hyperparameters tuned on BF16) and scored by ARI against species labels.
version: Table 3
metric: inference-throughput
limitations: Author-reported: two authors are at the DOE Joint Genome Institute, which develops GenomeOcean; no competing-interest statement is printed.; Microbial genomes and a metagenomic-binning fidelity check, not human diagnostic sequence; the paper says the fidelity metrics are benchmark-specific (Section 4.4).; Throughput is measured at one fixed batch size per model scale, which is not printed.; Peak VRAM is printed without spread; throughput and ARI are mean and standard deviation over five trials.; One GPU type (H100 SXM5); the paper says results for other GPUs and engines need their own evaluation.
memory measurement: torch.cuda.max_memory_allocated() (Section 2.6)
random seeds: 42-46
source locator: Sections 2.3-2.7; Table 3

Related records

uses data: GTDB genome fragments for metagenomic binning, 20 of a pre-sampled pool of 50 bacterial genomes per trial (Yu et al. 2026)
assessment: GenomeOcean-100M BF16 embedding extraction (Yu et al. 2026)
assessment: GenomeOcean-100M FP8 (E4M3) embedding extraction (Yu et al. 2026)
assessment: GenomeOcean-4B BF16 embedding extraction (Yu et al. 2026)
assessment: GenomeOcean-4B FP8 (E4M3) embedding extraction (Yu et al. 2026)
assessment: GenomeOcean-500M BF16 embedding extraction (Yu et al. 2026)
assessment: GenomeOcean-500M FP8 (E4M3) embedding extraction (Yu et al. 2026)
assessed by: Select an execution workflow for large-scale diagnostic genomics

Suggest a correction