0.0.26

pypi upload and timm fix

Support newer timm versions

0.0.25

version bump

Added PaSST-L weights

PaSST-L is the light version with fewer parameters, can be loaded like this:

from hear21passt.base import get_basic_model

model = get_basic_model(mode="logits", arch="passt_l_kd_p16_128_ap47")

Support loading different archs

from hear21passt.base import get_basic_model

model = get_basic_model(mode="logits", arch="passt_s_kd_p16_128_ap486")

0.0.21

Added a pre-trained model, trained with KD using the setup in https://github.com/fschmid56/EfficientAT

0.0.20

Added a pre-trained model, trained with KD using the setup in https://github.com/fschmid56/EfficientAT

0.0.19

• Fix broken links to pre-trained models
• Added readable error messages when pos-encoding shapes don't match.

0.0.18

Add support to 20 second and 30 second models.

0.0.17

Added pretrained model with no patch overlap PaSST-S-N on Audioset passt_s_p16_s16_128_ap468

0.0.16

update to 0.0.16