# How to publish this repo to Hugging Face

The challenge requires a PUBLIC Hugging Face repo containing the frozen weights, a runnable
inference entry point, pinned requirements, and the exact submission.csv. Everything in this
folder is that package. Follow these steps once (about 5 minutes).

## 1. One-time setup
- Make a free account at https://huggingface.co (if you do not have one).
- Get a WRITE access token: https://huggingface.co/settings/tokens  ->  New token  ->  role "Write".

## 2. Install the client  (already done: huggingface_hub 1.31.0, command is `hf`)
```
pip install -U "huggingface_hub[cli]"
```

## 3. Log in (do this yourself, it is interactive)
In the Claude Code prompt, type this with the leading `!` so it runs in your session:
```
! hf auth login
```
Paste your write token when asked. If it says "command not found", use `! ~/.local/bin/hf auth login`.

## 4. Create the repo and upload
Replace YOURNAME with your Hugging Face username:
```
! hf repos create YOURNAME/tartan-imu-velocity --repo-type model
! cd /home/ahmed2000/tartan_imu/release && hf upload YOURNAME/tartan-imu-velocity . . --repo-type model
```
That uploads infer.py, weights/, submission.csv, requirements.txt, and README.md.

## 5. Confirm it is public
Open https://huggingface.co/YOURNAME/tartan-imu-velocity in a browser. Settings must show
the repo as Public (it is public by default).

## 6. What to put in the Google Form
- The Hugging Face repo URL above.
- Attach submission.csv (also in this folder).
- Paste the method description (the abstract we wrote).
- Attach the technical report PDF (in your Downloads).

## Verify before you submit
Re-running inference must reproduce submission.csv. From this folder:
```
python infer.py --test_dir /home/ahmed2000/tartan_imu/test \
  --index /home/ahmed2000/tartan_imu/index/test_windows.csv --out /tmp/check.csv
```
Then compare /tmp/check.csv to submission.csv (already verified equal to 1e-6 here).