Add VQ-VAE and D3PM trained weights

README.md

@@ -1,3 +1,16 @@
+## Discrete generative models
+
+Trained model weights for:
+
+- VQ-VAE
+  - MNIST
+  - CIFAR-10
+- D3PM trained on corresponding VQ-VAE latent space
+  - MNIST / CIFAR-10
+  - uniform and absorbing-state kernels
+
+All weights are provided as `.pth` files.
+
 ---
 license: mit
 ---

d3pm/cifar10/uniform/d3pm_interrupt_step_443513_FINAL_CIFAR10_UNIFORM.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4ee226bd149e5351053aca6e982b00af6c27606d65f570c8341d8b5aa5b5088
+size 1409626227

d3pm/mnist/absorbing/d3pm_final_step_94000_MNIST_ABSORBING.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b699202a79b0cd04782945d37fb1985116f93862551e6da57b235df6098717bf
+size 242210443

d3pm/mnist/uniform/d3pm_final_step_94000_MNIST_UNIFORM.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:872a6ac0c5d86e66ddbf61a69a4cad2a7ad0d13d4c8e99ebef0129682fe72a53
+size 242210443

vqvae/cifar10/vqvae-cifar10.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e8cdda084ede89d6cbf37f49fa8686ff9a10bcb940a59909e33be8df1024699
+size 3001767

vqvae/mnist/vqvae-mnist-epoch=98-val_loss=0.013.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb233dc07e525ce29e84297c3dfe9187de7ade71e7d6d66874f070e72f3be9c4
+size 2985812