`Embedding`

A lookup table mapping integer indices to dense vectors.

class Embedding : Module— python/vkml/nn.py:604

Implemented as an index_select over the weight, so its gradient is a scatter_add back into a zero tensor — which is why scatter_add's determinism matters for training a model with an embedding, not only for inference.

Construction¶

`init`¶

def __init__self, num_embeddings: int, embedding_dim: int— python/vkml/nn.py:612

Forward¶

`forward`¶

def forwardself, indices: V.Tensor -> V.Tensor— python/vkml/nn.py:620

Internals¶

`repr`¶

def __repr__self -> str— python/vkml/nn.py:628

Embedding