metadata
license: mit
tags:
- generated_from_trainer
datasets:
- squad
model-index:
- name: squad_roberta_base
results: []
squad_roberta_base
This model is a fine-tuned version of roberta-base on the squad dataset.
Training and evaluation data
Trained and evaluated on the squad dataset.
Training procedure
Trained on 16 Graphcore Mk2 IPUs using optimum-graphcore.
Command line:
python examples/question-answering/run_qa.py \
--ipu_config_name Graphcore/roberta-base-ipu \
--model_name_or_path roberta-base \
--dataset_name squad \
--do_train \
--do_eval \
--num_train_epochs 2 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 2 \
--pod_type pod16 \
--learning_rate 6e-5 \
--max_seq_length 384 \
--doc_stride 128 \
--seed 1984 \
--lr_scheduler_type linear \
--loss_scaling 64 \
--weight_decay 0.01 \
--warmup_ratio 0.25 \
--logging_steps 1 \
--save_steps -1 \
--dataloader_num_workers 64 \
--output_dir squad_roberta_base \
--overwrite_output_dir \
--push_to_hub
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 6e-05
- train_batch_size: 4
- eval_batch_size: 2
- seed: 1984
- distributed_type: IPU
- total_train_batch_size: 256
- total_eval_batch_size: 40
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.25
- num_epochs: 2.0
- training precision: Mixed Precision
Training results
***** train metrics *****
epoch = 2.0
train_loss = 1.2528
train_runtime = 0:02:14.50
train_samples = 88568
train_samples_per_second = 1316.952
train_steps_per_second = 5.13
***** eval metrics *****
epoch = 2.0
eval_exact_match = 85.2696
eval_f1 = 91.7455
eval_samples = 10790
Framework versions
- Transformers 4.18.0.dev0
- Pytorch 1.10.0+cpu
- Datasets 2.0.0
- Tokenizers 0.11.6