MoRAM

This directory contains the code for evaluating MoRAM on the X-TAIL benchmark: continual few-shot adaptation of CLIP across 10 image classification domains.

Back to main README

---

Environment Setup

conda create -n moram_clip python=3.12 -y
conda activate moram_clip

# Install PyTorch matching your CUDA version, e.g.:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

# Install project dependencies
pip install -r requirements.txt

Data Preparation

The X-TAIL benchmark uses 10 classification datasets in the following default task sequence:

1. Aircraft
2. Caltech-101
3. DTD
4. EuroSAT
5. Oxford Flowers
6. Food-101
7. MNIST
8. Oxford Pets
9. Stanford Cars
10. SUN397

Download and organize all datasets under a single root directory. Follow the dataset preparation guide from CoOp DATASETS.md.

Your data directory should look like:

<data_dir>/
├── fgvc_aircraft/
├── caltech-101/
├── dtd/
├── eurosat/
├── oxford_flowers/
├── food-101/
├── mnist/
├── oxford_pets/
├── stanford_cars/
└── sun397/

Running MoRAM

Quick Start

From this directory, with your Python environment activated and datasets under XTAIL_DATA_DIR (default ./datasets):

cd XTAIL
bash runner_moram.sh

Acknowledgement

This benchmark builds on MoE-Adapters, RAIL, and CoDyRA.