diff --git a/examples/mnist/data/in/images b/examples/mnist/data/in/images
deleted file mode 100644
index bbce276..0000000
Binary files a/examples/mnist/data/in/images and /dev/null differ
diff --git a/examples/mnist/data/in/labels b/examples/mnist/data/in/labels
deleted file mode 100644
index d6b4c5d..0000000
Binary files a/examples/mnist/data/in/labels and /dev/null differ
diff --git a/examples/mnist/requirements.txt b/examples/mnist/requirements.txt
deleted file mode 100644
index 61aa5df..0000000
--- a/examples/mnist/requirements.txt
+++ /dev/null
@@ -1,4 +0,0 @@
-einops==0.6.1
-torch==2.0.0
-tqdm==4.64.1
-typing_extensions==4.6.3
diff --git a/examples/mnist/src/main.py b/examples/mnist/src/main.py
deleted file mode 100644
index 298fa07..0000000
--- a/examples/mnist/src/main.py
+++ /dev/null
@@ -1,48 +0,0 @@
-import warnings
-from pathlib import Path
-
-import torch
-from einops import rearrange
-from torchvision.utils import save_image
-
-import diffusion
-from diffusion.data import Identity
-from diffusion.loss import Simple
-from diffusion.net import Transformer
-from diffusion.noise import Normal
-from diffusion.schedule import Cosine
-
-dataset = "mnist"
-input, output = Path("data/in") / dataset, Path("data/out") / dataset
-output.mkdir(exist_ok=True, parents=True)
-torch.set_float32_matmul_precision('high')
-torch.set_grad_enabled(False)
-
-c, h, w, p, q = 1, 28, 28, 2, 2
-x = (input / "images").read_bytes()[16:]
-with warnings.catch_warnings():
-    warnings.simplefilter("ignore")
-    x = torch.frombuffer(x, dtype=torch.uint8).view(-1, c, h, w)
-x = rearrange(x / 127.5 - 1, "b c (h p) (w q) -> b (h w) (c p q)", p=p, q=q)
-y = torch.frombuffer((input / "labels").read_bytes()[8:], dtype=torch.uint8)
-
-model = diffusion.Model(
-    data=Identity(x, y, batch=128, shuffle=True),
-    schedule=Cosine(1000),
-    loss=Simple(),
-    noise=Normal(parameter="x", variance="fixed"),
-    net=Transformer(
-        input=x.shape[2],
-        labels=10,
-        hidden=768,
-        heads=12,
-        depth=12,
-    ),
-    device=torch.device("cuda"),
-)
-
-for epoch, loss in enumerate(model.train(epochs=100)):
-    l = model.sample(torch.arange(10))
-    l = (l[torch.linspace(0, l.shape[0] - 1, 10).long()] + 1) / 2
-    l = rearrange(l, "t b (h w) (c p q) -> c (b h p) (t w q)", h=h // p, p=p, q=q)
-    save_image(l, output / f"{epoch} ({loss:.2e}).png")
\ No newline at end of file