James Briggs
Added all files from training
df2d5cd
|
raw
history blame
2.1 kB
metadata
tags:
  - generated_from_trainer
datasets:
  - squad
model-index:
  - name: Graphcore/bert-base-uncased-squad
    results: []

Graphcore/bert-base-uncased-squad11

This model is a fine-tuned version of Graphcore/bert-base-uncased on the squad dataset.

Intended uses & limitations

More information needed

Training and evaluation data

squad dataset

Training procedure

Model was trained on 16 Graphcore Mk2 IPUs using the optimum-graphcore library.

Command line:

python examples/question-answering/run_qa.py \
  --model_name_or_path Graphcore/bert-base-uncased \
  --ipu_config_name Graphcore/bert-base-ipu \
  --dataset_name squad \
  --do_train \
  --do_eval \
  --num_train_epochs 3 \
  --per_device_train_batch_size 2 \
  --per_device_eval_batch_size 2 \
  --gradient_accumulation_steps 16 \
  --pod_type pod16 \
  --learning_rate 9e-5 \
  --max_seq_length 384 \
  --doc_stride 128 \
  --seed 42\
  --lr_scheduler_type linear \
  --loss_scaling 64 \
  --weight_decay 0.01 \
  --warmup_ratio 0.2 \
  --logging_steps 1 \
  --save_steps 50 \
  --dataloader_num_workers 64 \
  --ipu_config_overrides "embedding_serialization_factor=2" \
  --output_dir squad_v2_bert_base \
  --overwrite_output_dir

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 6e-05
  • train_batch_size: 2
  • eval_batch_size: 2
  • seed: 42
  • distributed_type: IPU
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 256
  • total_eval_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 3.0
  • training precision: Mixed Precision

Training results

{ 
    "epoch": 3.0,
    "eval_exact_match": 81.79754020813624,
    "eval_f1": 88.84840994541061,
    "eval_samples": 10784
}

Framework versions

  • Transformers 4.18.0.dev0
  • Pytorch 1.10.0+cpu
  • Datasets 1.18.4
  • Tokenizers 0.11.6