WJEPA Checkpoints

This repository contains WJEPA checkpoints from a single seed for six datasets. Each dataset has two checkpoint variants: participant and random.

These checkpoints accompany the W-JEPA code repository, which contains the model implementation, training pipeline, and inference scripts. See its README for setup and usage instructions.

There are 12 checkpoint files in total, approximately 744 MB.

Folder structure

.
β”œβ”€β”€ README.md
β”œβ”€β”€ dsads/
β”‚   β”œβ”€β”€ participant/latest_checkpoint.pth
β”‚   └── random/latest_checkpoint.pth
β”œβ”€β”€ mhealth/
β”‚   β”œβ”€β”€ participant/latest_checkpoint.pth
β”‚   └── random/latest_checkpoint.pth
β”œβ”€β”€ pamap2/
β”‚   β”œβ”€β”€ participant/latest_checkpoint.pth
β”‚   └── random/latest_checkpoint.pth
β”œβ”€β”€ realworld/
β”‚   β”œβ”€β”€ participant/latest_checkpoint.pth
β”‚   └── random/latest_checkpoint.pth
β”œβ”€β”€ uci_har/
β”‚   β”œβ”€β”€ participant/latest_checkpoint.pth
β”‚   └── random/latest_checkpoint.pth
└── usc_had/
    β”œβ”€β”€ participant/latest_checkpoint.pth
    └── random/latest_checkpoint.pth

Checkpoint paths follow <dataset>/<variant>/latest_checkpoint.pth. Select the dataset folder and either the participant or random variant to download the corresponding weights.

Download the weights

Install and authenticate

Install the Hugging Face Hub package, which includes the hf command:

python -m pip install -U huggingface_hub

For private repository access, log in with a Hugging Face account that has access to this repository. If using an access token, it needs read permission for zou-research-lab/wjepa.

hf auth login

Download all checkpoints

hf download zou-research-lab/wjepa \
  --include "*.pth" \
  --local-dir ./wjepa

This saves all 12 checkpoints under ./wjepa/, preserving the dataset and variant folders.

Download one dataset

For example, download both UCI HAR variants:

hf download zou-research-lab/wjepa \
  --include "uci_har/*/*.pth" \
  --local-dir ./wjepa

Replace uci_har with any dataset folder listed above.

Download one checkpoint

For example, download the UCI HAR participant checkpoint:

hf download zou-research-lab/wjepa \
  uci_har/participant/latest_checkpoint.pth \
  --local-dir ./wjepa

The file will be saved at ./wjepa/uci_har/participant/latest_checkpoint.pth.

Download with Python

After authenticating as described above, download a checkpoint with hf_hub_download:

from huggingface_hub import hf_hub_download

checkpoint_path = hf_hub_download(
    repo_id="zou-research-lab/wjepa",
    filename="uci_har/participant/latest_checkpoint.pth",
    local_dir="./wjepa",
)

print(checkpoint_path)

To download all checkpoints from Python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="zou-research-lab/wjepa",
    allow_patterns="*.pth",
    local_dir="./wjepa",
)
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support