diff --git a/src/lightning_trainable/datasets/toy/gmm.py b/src/lightning_trainable/datasets/toy/gmm.py
new file mode 100644
index 0000000..fd89191
--- /dev/null
+++ b/src/lightning_trainable/datasets/toy/gmm.py
@@ -0,0 +1,21 @@
+import torch
+import torch.distributions as D
+
+from ..core.distribution_dataset import DistributionDataset
+
+
+class GaussianMixtureModelDistribution(D.MixtureSameFamily):
+    """
+    Distribution that allows for multiple gaussian blobs with different means, stddevs and weights.
+
+    This behaviour is roughly equivalent to sklearn's make_blobs, but with control over the placement of the means in
+    comparison to the hyperspheres.
+    """
+
+    def __init__(self, means, stddevs, weights):
+        super().__init__(D.Categorical(weights), D.MultivariateNormal(means, torch.diag_embed(stddevs)))
+
+
+class GaussianMixtureModelDataset(DistributionDataset):
+    def __init__(self, means, stddevs, weights, **kwargs):
+        super().__init__(GaussianMixtureModelDistribution(means, stddevs, weights), **kwargs)