What does the "Percent" column in `prompts.csv` represent?

[eisrael123]

What does the "Percent" column represent? Is it:

• Data quality/completeness score from the source dataset?
• Sequencing coverage depth?
• Percentage of expected sequence that is present?
• Something else entirely?

I'd like to understand this metadata to better interpret the test results. Any clarification would be greatly appreciated!

Related Files

• evo2/test/data/prompts.csv
• evo2/test/test_evo2_generation.py (uses this file)
• evo2/notebooks/generation/generation_notebook.ipynb (references this file)

[garykbrixi]

Hi @eisrael123, the percent column is not used. Please see the test folder to see how these sequences are used. For generation a part is passed as prompt and rest is evaluated, for forward pass the whole sequence is used. This is primarily for testing correctness such as when deploying on different hardware