bert-l-squadv1.1-sl256

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

run01-bert-l-uwwm-squadv1.1-sl256-ds128-e2-tbs16

This model is a fine-tuned version of bert-large-uncased-whole-word-masking on the squad dataset. ONNX and OpenVINO IR are enclosed here.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

NEPOCH=2
TBS=16
EBS=64
SL=256
DS=128

cmd="
python run_qa.py \
    --model_name_or_path ${BASEM} \
    --dataset_name squad \
    --do_eval \
    --do_train \
    --evaluation_strategy steps \
    --eval_steps 500 \
    --learning_rate 3e-5 \
    --fp16 \
    --num_train_epochs $NEPOCH \
    --per_device_eval_batch_size $EBS \
    --per_device_train_batch_size $TBS \
    --max_seq_length $SL \
    --doc_stride $DS \
    --save_steps 1000 \
    --logging_steps 1 \
    --overwrite_output_dir \
    --run_name $RUNID \
    --output_dir $OUTDIR
"

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2.0
• mixed_precision_training: Native AMP

Training results

Best checkpoint was at step 11500 but it was not saved. This is final checkpoint (12K+).

  eval_exact_match = 86.9347
  eval_f1          = 93.1359
  eval_samples     =   12097

Framework versions

• Transformers 4.18.0
• Pytorch 1.11.0+cu113
• Datasets 2.1.0
• Tokenizers 0.12.1