README.md

metadata

license: apache-2.0
tags:
  - stripedhyena
  - long context
  - deep signal processing
  - hybrid
  - biology
  - genomics
task_categories:
  - text-generation
language:
  - en
pretty_name: open-genome
configs:
  - config_name: stage1
    data_files:
      - split: train
        path:
          - stage1/gtdb/gtdb_train_shard_*
          - stage1/imgpr/imgpr_train.parquet
      - split: validation
        path:
          - stage1/gtdb/gtdb_valid_small.parquet
          - stage1/imgpr/imgpr_valid_small.parquet
      - split: test
        path:
          - stage1/gtdb/gtdb_test.parquet
          - stage1/imgpr/imgpr_test.parquet
  - config_name: stage2
    data_files:
      - split: train
        path: stage2/train_stage2.parquet
      - split: validation
        path: stage2/valid_stage2.parquet
      - split: test
        path: stage2/test_stage2.parquet
  - config_name: sample
    data_files:
      - split: validation
        path: stage2/valid_stage2.parquet

Dataset organization

The OpenGenome dataset is organized in 2 stages, where stage 1 has context length 8k and stage 2 has context length 131k. Each stage has their own datasplits.

- stage1
  - train
  - validation
  - test

- stage2
  - train
  - validation
  - test

Instructions to download

You can load a dataset using HF's API, with an example below.

from datasets import load_dataset

stage1_data = load_dataset("LongSafari/open-genome", 'stage1')

# access just the train data
stage_1_train_data = stage1_data['train']

Note: stage 1 training dataset is sharded into separate files due to it's large size.

We also provide a small dataset sample to test out the pipeline if you prefer.

sample_data = load_dataset("LongSafari/open-genome", 'sample')['validation']