v0.8.2-p0

• Support for models with shared Embedding layers

v0.8.1-p0

• Allow conversion of models directly from memory to json, without saving as h5 in between
• Add support for LSTM layers with return_state=True
• Fix show_tensor5

v0.8.0-p0

• Improved error messages
• Fixed certain cases of Concatenate layer with non-default axis
• Support custom layers (by injecting user-land factory functions into fdeep::load_model)

v0.7.11-p0

• New --no-tests command-line flag for convert_model.py to optionally avoid generating tests at all.

v0.7.10-p0

• Support for CuDNNLSTM layers

v0.7.9-p0

• Support for CuDNNGRU layers
• Minor sanity checks
• New FAQ

v0.7.8-p0

• Add support for pooling layers with argument "channels_first"

v0.7.7-p0

• Support for conversion of YOLO models
• Support for models with multiple embedding layers with different input_dims

v0.7.6-p0

Support for AlphaDropout, GaussianDropout SpatialDropout1D, SpatialDropout2D and SpatialDropout3D.

v0.7.5-p0

• Add support for standalone Softmax layer.
• Fix import of standalone ReLU layer without maximum value.