mlx_neural_engine (mlx v0.2.0)
View SourceMLX Neural Engine - INSANE AI/ML Pipeline for Apple Silicon This module implements a complete neural network training and inference engine using MLX with Erlang's fault-tolerance and concurrency
Summary
Functions
Batch normalization
2D Convolution operation
Create a neural network with specified architecture
Distributed SGD training
Dropout for regularization
Make predictions with a trained network
Streaming inference for real-time applications
Train a neural network
Transformer attention mechanism
Functions
Batch normalization
-spec code_change(term(), #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}, term()) -> {ok, #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}}.
-spec convolution_2d(Input :: reference(), Kernel :: reference(), Options :: map()) -> {ok, reference()}.
2D Convolution operation
-spec create_network(Architecture :: map(), Options :: map()) -> {ok, reference()} | {error, term()}.
Create a neural network with specified architecture
Distributed SGD training
Dropout for regularization
-spec handle_call(term(), {pid(), term()}, #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}) -> {reply, term(), #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}}.
-spec handle_cast(term(), #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}) -> {noreply, #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}}.
-spec handle_info(term(), #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}) -> {noreply, #state{networks :: term(), optimizers :: term(), training_jobs :: term(), inference_servers :: term(), distributed_nodes :: term()}}.
Make predictions with a trained network
-spec stop() -> ok.
-spec streaming_inference(NetworkRef :: reference(), InputStream :: reference()) -> {ok, reference()}.
Streaming inference for real-time applications
-spec train_network(NetworkRef :: reference(), TrainingData :: term(), Options :: map()) -> {ok, map()} | {error, term()}.
Train a neural network
-spec transformer_attention(Query :: reference(), Key :: reference(), Value :: reference()) -> {ok, reference()}.
Transformer attention mechanism