API Reference: functional

grilly.functional provides stateless operations that mirror torch.nn.functional.

Activations

Function	Signature	Description
`linear`	`linear(x, weight, bias=None)`	Affine transformation: `x @ weight.T + bias`.

Function	Signature	Description
`attention`	`attention(q, k, v, mask=None)`	Standard scaled dot-product attention.
`flash_attention2`	`flash_attention2(q, k, v)`	Memory-efficient Flash Attention 2.

Function	Signature	Description
`layer_norm`	`layer_norm(x, weight, bias, eps=1e-5)`	Layer normalization.

Function	Signature	Description
`dropout`	`dropout(x, p=0.5, training=True)`	Randomly zero elements.

Function	Signature	Description
`cross_entropy`	`cross_entropy(logits, targets)`	Cross-entropy loss.
`binary_cross_entropy`	`binary_cross_entropy(predictions, targets)`	Binary cross-entropy.

Function	Signature	Description
`memory_read`	`memory_read(memory, query)`	Read from memory via attention.
`memory_write`	`memory_write(memory, key, value)`	Write to memory.
`memory_context_aggregate`	`memory_context_aggregate(contexts)`	Aggregate memory contexts.
`memory_query_pooling`	`memory_query_pooling(queries)`	Pool memory queries.
`memory_inject_concat`	`memory_inject_concat(x, memory)`	Inject memory via concatenation.
`memory_inject_gate`	`memory_inject_gate(x, memory)`	Inject memory via gating.
`memory_inject_residual`	`memory_inject_residual(x, memory)`	Inject memory via residual.

Function	Signature	Description
`place_cell`	`place_cell(position, centers)`	Place cell activation (spatial).
`time_cell`	`time_cell(time, centers)`	Time cell activation (temporal).
`theta_gamma_encoding`	`theta_gamma_encoding(phase, frequency)`	Theta-gamma neural encoding.

Function	Signature	Description
`fisher_info`	`fisher_info(grads)`	Compute Fisher information matrix.
`ewc_penalty`	`ewc_penalty(params, fisher, old_params)`	Elastic Weight Consolidation penalty.
`natural_gradient`	`natural_gradient(grad, fisher)`	Natural gradient via Fisher.
`nlms_predict`	`nlms_predict(weights, x)`	NLMS prediction.
`nlms_update`	`nlms_update(weights, x, error, mu)`	NLMS weight update.
`whitening_transform`	`whitening_transform(x)`	Compute whitening transform.
`whitening_apply`	`whitening_apply(x, transform)`	Apply whitening.

Function	Signature	Description
`continuous_to_spikes`	`continuous_to_spikes(x, threshold)`	Convert continuous signal to spikes.
`spikes_to_continuous`	`spikes_to_continuous(spikes, decay)`	Decode spikes to continuous signal.
`bridge_temporal_weights`	`bridge_temporal_weights(t, decay)`	Temporal weighting for bridge.

Function	Signature	Description
`embedding_lookup`	`embedding_lookup(table, indices)`	Lookup embeddings by index.
`embedding_normalize`	`embedding_normalize(embeddings)`	L2-normalize embeddings.
`embedding_position`	`embedding_position(embeddings, positions)`	Add positional embeddings.
`embedding_pool`	`embedding_pool(embeddings, method)`	Pool embeddings.
`embedding_ffn`	`embedding_ffn(embeddings, weights)`	FFN on embeddings.
`embedding_attention`	`embedding_attention(q, k, v)`	Attention over embeddings.

Function	Signature	Description
`faiss_distance`	`faiss_distance(queries, keys)`	Compute distances.
`faiss_topk`	`faiss_topk(distances, k)`	Top-k selection.
`faiss_ivf_filter`	`faiss_ivf_filter(queries, centroids)`	IVF coarse filtering.
`faiss_kmeans_update`	`faiss_kmeans_update(centroids, data)`	K-means centroid update.
`faiss_quantize`	`faiss_quantize(data, codebook)`	Product quantization.

Function	Signature	Description
`lif_step`	`lif_step(x, v, tau, v_threshold)`	Single LIF neuron step. Returns `(spike, v_new)`.
`if_step`	`if_step(x, v, v_threshold)`	Single IF neuron step.
`multi_step_forward`	`multi_step_forward(module, x, T)`	Run module over T time steps.
`seq_to_ann_forward`	`seq_to_ann_forward(module, x)`	Apply ANN module to temporal data.
`reset_net`	`reset_net(module)`	Reset all neuron membrane potentials.
`set_step_mode`	`set_step_mode(module, mode)`	Set single-step or multi-step mode.