Model card for boldgpt_small_patch10.cont

A Vision Transformer (ViT) model trained on BOLD activation maps from NSD-Flat. The training objective was to auto-regressively predict the next patch with shuffled patch order and MSE loss. This model was trained using shared1000 as the held out validation set.

Dependencies

• boldGPT

Usage

from boldgpt.data import ActivityTransform
from boldgpt.models import create_model
from datasets import load_dataset

model = create_model("boldgpt_small_patch10.cont", pretrained=True)

dataset = load_dataset("clane9/NSD-Flat", split="train")
dataset.set_format("torch")

transform = ActivityTransform()
batch = dataset[:1]
batch["activity"] = transform(batch["activity"])

# output: (B, N + 1, D) predicted next patches
output, state = model(batch)

Reproducing

• Training command:

  torchrun --standalone --nproc_per_node=4 \
    scripts/train.py \
    --out_dir results \
    --model boldgpt_small_patch10 \
    --no_cat --shuffle --epochs 1000 --bs 512 \
    --workers 0 --amp --compile --wandb
  

• Commit: e0b29adc8d5b3ed2f1a555d7de4754ba96a3bb3e