CLI Reference

This page lists all available command-line parameters for the OmniGenBench (ogb) toolkit.
For usage examples, see the Command-Line Usage page.

---

Top-Level Command: ogb

The main CLI entry point.
Available subcommands:

• autoinfer — Run inference with fine-tuned models
• autotrain — Train or fine-tune genomic models
• autobench — Benchmark models on curated datasets
• rna_design — Design RNA sequences for target secondary structures

To view global options:

ogb --help

---

ogb autoinfer

Run inference with fine-tuned genomic foundation models.

Parameters

Parameter	Type	Required	Default	Description
--model	str	✓	—	Model name or path (e.g., yangheng/ogb_tfb_finetuned).
--sequence	str	No	—	Single sequence or comma-separated sequences.
--input-file	str	No	—	JSON/CSV/TXT file containing sequences.
--output-file	str	No	inference_results.json	Output JSON file.
--batch-size	int	No	32	Inference batch size.
--device	str	No	auto	Device to run inference (cuda:0, cpu).

Note

At least one of --sequence or --input-file must be provided.

Example

ogb autoinfer \
  --model yangheng/ogb_tfb_finetuned \
  --sequence "ATCGATCGATCGATCGATCGATCGATCGATCG" \
  --output-file tfb_predictions.json

---

ogb autotrain

Automatically fine-tune genomic foundation models.

Parameters

Parameter	Type	Required	Default	Description
--dataset, -d	str	✓	—	Dataset name or path.
--model, -m	str	✓	—	Base model name or path.
--tokenizer	str	No	model default	Tokenizer to use.
--output-dir	str	No	—	Directory to save fine-tuned model.
--num-epochs	int	No	—	Training epochs.
--batch-size	int	No	—	Training batch size.
--learning-rate	float	No	—	Learning rate.
--overwrite	flag	No	—	Overwrite output directory.
--trainer	str	No	accelerate	Trainer backend.

Example

ogb autotrain \
  --dataset yangheng/tfb_promoters \
  --model zhihan1996/DNABERT-2-117M \
  --num-epochs 10

---

ogb autobench

Benchmark genomic models on standard datasets.

Parameters

Parameter	Type	Required	Default	Description
--model, -m	str	✓	—	Model to benchmark.
--benchmark, -b	str	✓	—	Benchmark dataset (RGB, GUE, PGB, BEACON).
--tokenizer, -t	str	No	model default	Tokenizer to use.
--trainer	str	No	accelerate	Trainer backend.
--overwrite	flag	No	—	Overwrite existing benchmark results.

Example

ogb autobench \
  --model yangheng/OmniGenome-186M \
  --benchmark RGB

---

ogb rna_design

Design RNA sequences for target secondary structures.

Parameters

Parameter	Type	Required	Default	Description
--structure	str	✓	—	Target structure in dot-bracket notation.
--model	str	No	yangheng/OmniGenome-186M	Model for MLM-guided mutation.
--mutation-ratio	float	No	0.1	Fraction of nucleotides mutated each generation.
--num-population	int	No	100	GA population size.
--num-generation	int	No	100	Maximum generations.
--output-file	str	No	—	Save final sequences to file.

Example

ogb rna_design \
  --structure "(((...)))" \
  --output-file designs.txt