fix 37

[antoine-tran]

Why ?

fixes #37

The PR fixes the renaming issues in checkpoint compilation script.

Test plan

1. Train a new dummy model using AudioCraft:

git clone git@github.com:facebookresearch/audiocraft.git
cd audiocraft
python -m pip install -e ".[wm]"

AUDIOCRAFT_DORA_DIR="/tmp/wm_${USER}" \
    dora run device=cpu dataset.num_workers=0 optim.epochs=1 \
    dataset.train.num_samples=10 \
    dataset.valid.num_samples=10 \
    dataset.evaluate.num_samples=10 \
    dataset.generate.num_samples=10 \
    solver=watermark/robustness \
    checkpoint.save_last=true \
    dset=audio/example

/tmp/wm_${USER}/xps/SOME_HASH_CODE/checkpoint.th

You should see the output such as:

2. Generate the checkpoints in audioseal-friendly format:

cd [AudioSeal code directory]
python src/scripts/checkpoints.py /tmp/wm_${USER}/xps/SOME_HASH_CODE/checkpoint.th OUTPUT_DIR

3. Try load the checkpoints using AudioSeal API:

from audioseal import AudioSeal
generator = AudioSeal.load_generator("[OUTPUT_DIR/checkpoint_generator_base.pth", nbits=16)

detector = AudioSeal.load_detector("[OUTPUT_DIR/checkpoint_detector_base.pth", nbits=16)