MAGICS-LAB/DNABERT_2 — reverse-engineered prompt

I want this repo turned into a clean, runnable DNABERT 2 project that I can actually use without digging through papers. Make it easy to load the pretrained model, paste in one or more DNA sequences, and get back embeddings in a simple way. I also want a straightforward fine tuning flow for my own dataset using train.csv, dev.csv, and test.csv in the sample format, with clear commands and sensible defaults.

If the benchmark dataset is available, add an easy way to run the provided evaluation scripts too, but the main goal is inference and fine tuning on custom data. Please make the setup smooth, check dependencies, and add a simple example script or small CLI so I can test everything end to end quickly. If there are places where the repo assumes multiple GPUs, make it still work on a single GPU or CPU when possible. Clean up anything obviously broken, and update the docs so a biology person with light coding experience can follow it. Look up current docs online if you need to.