Why after standard lstm fine tuning initializing on reduced character set with OEM_TESSERACT_LSTM_COMBINED reports error, and trhe only valid mode is OEM_LSTM_ONLY ?
Is there any way to keep combined mode working, or if not - how does it affect the recognition quality ?