********** Layers API ********** Layers are the basic building blocks of neural networks in FlexFlow. The inputs of a layer consists of a tensor or a list of tensors and some state variables, and the outputs of a layer is a tensor or a list of tensors. See https://github.com/flexflow/FlexFlow/examples/python/native/ops for an example for every layer .. automodule:: flexflow.core.flexflow_cffi :noindex: Conv2D ====== .. autoclass:: FFModel() :noindex: :members: conv2d Pool2D ====== .. autoclass:: FFModel() :noindex: :members: pool2d Dense ====== .. autoclass:: FFModel() :noindex: :members: dense Embedding ========= .. autoclass:: FFModel() :noindex: :members: embedding Transpose ========= .. autoclass:: FFModel() :noindex: :members: transpose Reverse ======= .. autoclass:: FFModel() :noindex: :members: reverse Concatenate =========== .. autoclass:: FFModel() :noindex: :members: concat Split ====== .. autoclass:: FFModel() :noindex: :members: split Reshape ======= .. autoclass:: FFModel() :noindex: :members: reshape Flat ====== .. autoclass:: FFModel() :noindex: :members: flat BatchNorm ========= .. autoclass:: FFModel() :noindex: :members: batch_norm BatchMatMul =========== .. autoclass:: FFModel() :noindex: :members: batch_matmul Add ====== .. autoclass:: FFModel() :noindex: :members: add Subtract ======== .. autoclass:: FFModel() :noindex: :members: subtract Multiply ======== .. autoclass:: FFModel() :noindex: :members: multiply Divide ====== .. autoclass:: FFModel() :noindex: :members: divide Exponential =========== .. autoclass:: FFModel() :noindex: :members: exp ReLU ==== .. autoclass:: FFModel() :noindex: :members: relu ELU ==== .. autoclass:: FFModel() :noindex: :members: elu Sigmoid ======= .. autoclass:: FFModel() :noindex: :members: sigmoid Tanh ==== .. autoclass:: FFModel() :noindex: :members: tanh Softmax ======= .. autoclass:: FFModel() :noindex: :members: softmax Dropout ======= .. autoclass:: FFModel() :noindex: :members: dropout MultiheadAttention ================== .. autoclass:: FFModel() :noindex: :members: multihead_attention