CTranslate2 1.5.0

CTranslate2 1.5.0

New features

• [C++] Add max_batch_size translation options for single translators

Fixes and improvements

• Improve INT8 performance on CPU
• Enable INT8 support on default Intel MKL build
• Simplify project dependencies:
  • Replace boost::program_options with cxxopts for client options
  • Include header-only dependencies as Git submodules (cxxopts, cub, and thrust)
  • Remove MKL-DNN
• Harmonize Python/C++ default values:
  • [Python] Change default beam size from 4 to 2
  • [C++] Load models on the CPU by default