RAME-VL

This repository hosts the released checkpoint for RAME-VL.

The checkpoint files expected by the accompanying codebase are:

• config.yaml
• model.pt

Clone This Repository

To clone the full repository with large files:

brew install git-xet
git xet install

git clone https://huggingface.co/Ritesh-hf/RAME-VL

To clone without downloading the checkpoint weights:

GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/Ritesh-hf/RAME-VL

Download Only the Checkpoint Files

If you only need the files required for evaluation:

huggingface-cli download Ritesh-hf/RAME-VL \
  config.yaml model.pt \
  --local-dir checkpoints/rame-vl

After download, the checkpoint directory should look like:

checkpoints/rame-vl/
  config.yaml
  model.pt

Using the Checkpoint in the Codebase

The accompanying codebase expects a local checkpoint directory containing config.yaml and model.pt.

Example evaluation command:

torchrun --standalone --nnodes=1 --nproc_per_node=1 \
  launch_scripts/eval_downstream.py \
  checkpoints/rame-vl \
  test-low-res \
  --device_batch_size 1 \
  --save_dir outputs/test_low_res \
  --overwrite

Notes

• This repository stores the released model checkpoint only.
• Datasets are not bundled in this repository.
• Some evaluation tasks in the codebase expect local dataset paths to be set up separately.