stanjsx/fMRI-LM-B-Qwen3-0.6B

fMRI-LM-B (Qwen3-0.6B) — pretrained backbones

Stage-1 tokenizers and stage-2 paired-pretraining checkpoints for fMRI-LM, a foundation model that aligns functional MRI with language.

• —Code: <https://github.com/yuxiangwei0808/fMRI-LM>
• —Paper: arXiv:2511.21760

Files

Three variants, differing in the objective the stage-1 fMRI tokenizer was trained with.

`vq-contrastive/` — vector quantization + SigLIP contrastive alignment.

`vq-domain/` — vector quantization + adversarial domain loss.

`mae/` — masked autoencoding (mask ratio 0.5) + adversarial domain loss.

All three were trained on UK Biobank with robust normalisation and Qwen3-0.6B. Stage-2 files are DeepSpeed checkpoints already merged to a single file.

The MAE stage-1 file loads with MaskedAutoencoderViT; the two VQ stage-1 files load with the Tokenizer class. They are not interchangeable.

Usage

Clone the repo, place a stage-1 file where stage 2 expects it, or a stage-2 file where stage 3 expects it, and run the corresponding script in scripts/. Stage 3 reads its parent through $STAGE2_CKPT:

export STAGE2_CKPT=/path/to/vq-contrastive/stage2-pretrain-Qwen3-0.6B.pt
bash scripts/launch_train_instruction.sh

Inputs must be preprocessed as the repo README describes: TR resampled to 2.0 s, 160 timepoints, 450 ROIs (Schaefer-400 + Tian-S3), then robust z-scored per ROI with site-wise variance normalisation.

Research use only.

Citation

@article{wei2025fmrilm,
  title   = {fMRI-LM: Towards a Universal Foundation Model for Language-Aligned fMRI Understanding},
  author  = {Wei, Yuxiang and Zhang, Yanteng and Xiao, Xi and Qian, Chengxuan and Wang, Tianyang and Calhoun, Vince D.},
  journal = {arXiv preprint arXiv:2511.21760},
  year    = {2025}
}