Skip to content

Latest commit

 

History

History
37 lines (25 loc) · 865 Bytes

README.md

File metadata and controls

37 lines (25 loc) · 865 Bytes
Raw

Finetuning SegaBERT

After pre-training, convert the model file to HuggingFace format with

PYTHONPATH=absolute_path_of(./transformers/) python ./transformers/transformers/convert_segatron_to_huggingface.py --segatron_path <your model file path> --bert_config_file <bert config file path> --huggingface_path <target path>

QQP MNLI QNLI

./transformers/examples/eval_finetune/train_glue.sh 0,1,2,3,4,5,6,7 16000

Grid search for MRPC RTE CoLA STS-B SST-2

./transformers/examples/eval_finetune/grid_search_glue.sh

SQUAD v1.1 or v2.0

./transformers/examples/eval_finetune/train_squad.sh 0,1,2,3,4,5,6,7 16000
./transformers/examples/eval_finetune/train_squad2.sh 0,1,2,3,4,5,6,7  16000

RACE

./transformers/examples/eval_finetune/run_race.sh 0,1,2,3,4,5,6,7 16000