ONNX Runtime optimizater ran on https://huggingface.co/SEBIS/code_trans_t5_base_commit_generation_transfer_learning_finetune. The fine-tune does well, the multi-task doesn't, small makes mistakes and large is 3Gb, so I settled on this one.

The intention is to run a local LLM to generate commit messages so OpenAI doesn't just get your IP for free.