CosineSimilarityLoss

Computes the cosine similarity between labels and predictions. Note: Keras defaults to negative similarity (loss).

Abstract Signature:

CosineSimilarityLoss(y_true: Tensor, y_pred: Tensor, axis: int = -1)

PyTorchJAX (Core)KerasTensorFlowApple MLXFlax NNXPaxML / Praxis

PyTorch

API: torch.nn.functional.cosine_similarity

Strategy: Macro '-torch.nn.functional.cosine_similarity({y_pred}, {y_true}, dim={axis})'

JAX (Core)

API: optax.cosine_similarity

Strategy: Macro '-optax.cosine_similarity({y_pred}, {y_true})'

Keras

API: keras.losses.cosine_similarity

Strategy: Direct Mapping

Official Docs ↗

TensorFlow

API: tf.keras.losses.cosine_similarity

Strategy: Direct Mapping

Official Docs ↗

Apple MLX

API: mlx.nn.losses.cosine_similarity

Strategy: Macro 'mlx.nn.losses.cosine_similarity({y_pred}, {y_true}, axis={axis})'

Flax NNX

API: optax.cosine_similarity

Strategy: Macro '-optax.cosine_similarity({y_pred}, {y_true})'

PaxML / Praxis

API: optax.cosine_similarity

Strategy: Macro '-optax.cosine_similarity({y_pred}, {y_true})'