Views
No views yet
vuiseng9/bert-l-squadv1.1-sl384 using OpenVINO NNCF.1# used 4xV100 GPUS
2# --fp16 for lower turnaround and resource requirement
3python run_qa.py \
4 --model_name_or_path bert-large-uncased-whole-word-masking-finetuned-squad \
5 --dataset_name squad \
6 --do_eval \
7 --do_train \
8 --evaluation_strategy steps \
9 --eval_steps 250 \
10 --learning_rate 3e-5 \
11 --fp16 \
12 --num_train_epochs 2 \
13 --per_device_eval_batch_size 64 \
14 --per_device_train_batch_size 8 \
15 --max_seq_length 384 \
16 --doc_stride 128 \
17 --save_steps 500 \
18 --logging_steps 1 \
19 --overwrite_output_dir \
20 --nncf_config nncf_bert_config_squad_kd.json \ #stock config which is also enclosed here
21 --run_name $RUNID \
22 --output_dir $OUTDIRvuiseng9/transformers (fork) , commit: ff24569b, NNCF v2.1+ commit (8e26365)1git clone https://huggingface.co/vuiseng9/nncf-qat-kd-bert-l-squadv1.1-sl384
2python run_qa.py \
3 --model_name_or_path ./nncf-qat-kd-bert-l-squadv1.1-sl384 \
4 --dataset_name squad \
5 --nncf_config nncf-qat-kd-bert-l-squadv1.1-sl384/nncf_bert_config_squad_kd.json \
6 --nncf_ckpt ./nncf-qat-kd-bert-l-squadv1.1-sl384 \
7 --do_eval \
8 --per_device_eval_batch_size 128 \
9 --max_seq_length 384 \
10 --doc_stride 128 \
11 --output_dir /tmp/eval-nncf-qat-kd-bert-l-squadv1.1-sl384 \
12 --overwrite_output_dir eval_exact_match = 87.1523
eval_f1 = 93.2668
eval_samples = 10784