Model description:

The Mistral-7B-v0.1, LoRA-tuned on RSG sets. For the information about inference see: https://github.com/IlyaGusev/rulm/blob/master/self_instruct/src/benchmarks/eval_lora_rsg.py train: https://github.com/IlyaGusev/rulm/blob/master/self_instruct/src/train.py The main config for LoRA: https://github.com/IlyaGusev/rulm/blob/master/self_instruct/configs/mistral_7b_rsg.json configs for separate tasks are in the same folder. The Mistral-7B-v01 model was trained on multi-task with the main config, merged into the main model, and then task-level LoRA adapters were trained on top of this merged model. zero-shot evaluation script: https://github.com/IlyaGusev/rulm/blob/master/self_instruct/src/benchmarks/eval_zs_rsg.py

Parameter description: