iMF and GAR Decoder Adaptation Checkpoints
This repository contains two groups of checkpoints.
Official iMF PyTorch checkpoints
These checkpoints were converted from the official JAX iMF checkpoints to PyTorch:
official-pytorch/iMF-B-2.ptofficial-pytorch/iMF-M-2.ptofficial-pytorch/iMF-L-2.ptofficial-pytorch/iMF-XL-2.pt
GAR decoder-adapted checkpoints
These checkpoints are improved decoder checkpoints from GAR decoder adaptation with random noise level in [0.25, 0.45]:
gar-decoder-noise025045/iMF-B-2_GAR_decoder_noise0.25-0.45_final.pth.targar-decoder-noise025045/iMF-M-2_GAR_decoder_noise0.25-0.45_final.pth.targar-decoder-noise025045/iMF-L-2_GAR_decoder_noise0.25-0.45_final.pth.targar-decoder-noise025045/iMF-XL-2_GAR_decoder_noise0.25-0.45_final.pth.tar
The GAR checkpoints are the final checkpoints from the evaluated epoch-10 runs. OpenAI-protocol gFID w/o CFG from the corresponding evaluated runs:
| model | gFID w/o CFG | rFID | IS |
|---|---|---|---|
| iMF-B-2 | 12.4139 | 0.8512 | 99.6011 |
| iMF-M-2 | 9.6555 | 0.7473 | 115.6061 |
| iMF-L-2 | 7.4068 | 0.7030 | 134.8000 |
| iMF-XL-2 | 7.6054 | 0.7180 | 138.1161 |
See manifest.json for file names and metadata.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support