Protocol

DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2)

Evidence concern: excluded from comparisons

Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4

Maximum batch size that fits in memory and mean throughput over 10 timing runs for nine encoder configurations.

9 evaluations · 92 results

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.

9 recorded evaluations, 92 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

9 evaluations · 92 results. Different protocols are not a single leaderboard.

Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) · Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)

Sorted by Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) (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	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 256 tokens)	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 512 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 256 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 512 tokens)	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens)	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens)	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 4096 tokens)	Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 8192 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 4096 tokens)	Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 8192 tokens)
ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)Evaluation	552 (best in column)	529 (best in column)	3631	1811	508 (best in column)	479 (best in column)	426 (best in column)	333 (best in column)	901 (best in column)	446 (best in column)	218 (best in column)	92 (best in column)
ModernGENA large (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)Evaluation	224	218	2721	1356	212	203	186	154	673	332	161	67
GENA base on one A100 80 GB (Aspidova et al. 2026)Evaluation	91.8	82	6637 (best in column)	2206 (best in column)	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
ModernGENA base on one A100 80 GB (Aspidova et al. 2026)Evaluation	78.1	77.7	3631	1811	77.2	76.4	74.8	71.2	901 (best in column)	446 (best in column)	218 (best in column)	92 (best in column)
NTv2 (100M) on one A100 80 GB (Aspidova et al. 2026)Evaluation	70.9	60.3	5941	1845	46.5	31.7	19	9.37	519	132	26	5
DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)Evaluation	68.6	62.6	3802	1433	53.4	40.7	27.6	15.5	376	90	15	5
NTv2 (250M) on one A100 80 GB (Aspidova et al. 2026)Evaluation	33.5	30.6	4107	1649	26.1	20.1	13.5	7.35	486	127	26	5
GENA large on one A100 80 GB (Aspidova et al. 2026)Evaluation	27.1	24.8	4922	1633	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
ModernGENA large on one A100 80 GB (Aspidova et al. 2026)Evaluation	26.2	26.1	2721	1356	26	25.9	25.6	24.9	673	332	161	67

All 92 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: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	376 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 1024, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	53.4 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 1024, column 'tokens/s'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	90 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 2048, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	40.7 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 2048, column 'tokens/s'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	3800 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 256, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	68.6 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 256, column 'tokens/s'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	15 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 4096 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 4096, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	27.6 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 4096 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 4096, column 'tokens/s'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	1430 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 512, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	62.6 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 512, column 'tokens/s'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	5 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 8192 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 8192, column 'bs'
Configuration: DNABERT-2 on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	15.5 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 8192 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source DNABERT-2 (117M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'DNABERT-2 (117M)', sequence length L = 8192, column 'tokens/s'
Configuration: GENA base on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	6640 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA base (110M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA base (110M)', sequence length L = 256, column 'bs'
Configuration: GENA base on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	91.8 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA base (110M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA base (110M)', sequence length L = 256, column 'tokens/s'
Configuration: GENA base on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	2210 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA base (110M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA base (110M)', sequence length L = 512, column 'bs'
Configuration: GENA base on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	82 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA base (110M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA base (110M)', sequence length L = 512, column 'tokens/s'
Configuration: GENA large on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	4920 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA large (336M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA large (336M)', sequence length L = 256, column 'bs'
Configuration: GENA large on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	27.1 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA large (336M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA large (336M)', sequence length L = 256, column 'tokens/s'
Configuration: GENA large on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	1630 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA large (336M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA large (336M)', sequence length L = 512, column 'bs'
Configuration: GENA large on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	24.8 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 512 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source GENA large (336M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'GENA large (336M)', sequence length L = 512, column 'tokens/s'
Configuration: ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	901 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source ModernGENA base (FA2 + autocast bf16) (135M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'ModernGENA base (FA2 + autocast bf16) (135M)', sequence length L = 1024, column 'bs'
Configuration: ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	508 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 1024 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source ModernGENA base (FA2 + autocast bf16) (135M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'ModernGENA base (FA2 + autocast bf16) (135M)', sequence length L = 1024, column 'tokens/s'
Configuration: ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	446 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source ModernGENA base (FA2 + autocast bf16) (135M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'ModernGENA base (FA2 + autocast bf16) (135M)', sequence length L = 2048, column 'bs'
Configuration: ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	479 Inference throughput (Encoder forward pass on one A100 80 GB; sequence length 2048 tokens) sample-per-second · higher Uncertainty: Not reported by the source: Mean of 10 timing runs; spread not printed Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source ModernGENA base (FA2 + autocast bf16) (135M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'ModernGENA base (FA2 + autocast bf16) (135M)', sequence length L = 2048, column 'tokens/s'
Configuration: ModernGENA base (FA2 + autocast bf16) on one A100 80 GB (Aspidova et al. 2026)	Protocol: DNA language model inference throughput and maximum batch size on one A100 80 GB by sequence length (Aspidova et al. 2026 Table 2) Dataset: Fixed-length token sequences of 256 to 8,192 tokens (Aspidova et al. 2026)	3630 Maximum batch size (Encoder forward pass on one A100 80 GB; sequence length 256 tokens) count · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix B Table 2 and Figure 1 (axis 'Inference speed (tokens/s)') give throughput in tokens/s, but the printed values are inconsistent with the printed batch sizes: ModernGENA base (FA2 + autocast bf16) at L = 256 prints 552.44 tokens/s with a batch of 3,631 sequences (929,536 tokens), about 28 minutes per batch, and ModernGENA base without FA2 at L = 8192 prints 71.15 tokens/s with a batch of 92 (753,664 tokens), about 2.9 hours per batch, against a stated pretraining throughput of 1.33M tokens/s on eight A100 GPUs (Section 3.2.4). The unit or scale of the throughput column is unclear. Batch sizes share the table. Values are recorded as printed.Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9); Figure 1 (page 4); Section 3.2.4 Methods, coverage and source ModernGENA base (FA2 + autocast bf16) (135M) inference throughput on one A100 (Aspidova et al. 2026) model-execution-20261011-protocol-aspidova2026-a100-inference-throughput Aggregation: Not reported Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Appendix B Table 2 (page 9), row 'ModernGENA base (FA2 + autocast bf16) (135M)', sequence length L = 256, column 'bs'

Source checking is not independent reproduction. Release 2026-10-23-5347695500ed.

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
External evaluations: 3

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

Null control

Proposed control: requires review

Unintegrated representation using protocol preprocessing

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

Protocol-specific conventional integration method

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 2026-10-23-5347695500ed. 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.

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 2026-10-23-5347695500ed
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 2026-10-23-5347695500ed · Record review: source checked

1 source record and release history

Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models · Original source · bioRxiv preprint version 1, posted 2026-04-23

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: model-execution-20261011-protocol-aspidova2026-a100-inference-throughput

areas: dna-genomes
contexts: research
protocol: For each sequence length L, the maximum batch size that fits in memory is found by exponential growth until an error, then binary search; throughput in tokens/s is averaged over 10 timing runs at that batch size on one NVIDIA A100 (80 GB).
version: Appendix B Table 2
metric: inference-throughput
limitations: Author-reported: the authors develop ModernGENA and GENA-LM; DNABERT-2 and NTv2 were run by them.; The printed tokens/s values are inconsistent with the printed batch sizes and the stated pretraining throughput (1.33M tokens/s on eight A100s, Section 3.2.4); the source carries an evidence concern, and values are kept as printed.; Inputs are fixed-length sequences whose content is not described; throughput does not depend on biology, but the input source is not stated.; Throughput is the mean of 10 timing runs; no spread is printed.; Models tokenize differently (BPE, 6-mer), so the same token length covers different numbers of base pairs.; GENA base and large print no values from 1,024 tokens upwards.
source locator: Section 3.4; Appendix B Table 2

Related records

Suggest a correction