Hugging Face's logo

Graphcore
/
roberta-base-squad2

Question Answering
Transformers
PyTorch
Graphcore
roberta
Generated from Trainer
Inference Endpoints
Model card Files Community
roberta-base-squad2
File size: 2,612 Bytes 
a11cf16
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
689344c
a11cf16
 
 
689344c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a11cf16
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
689344c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a11cf16
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
 
---
license: mit
tags:
- generated_from_trainer
datasets:
- squad_v2
model-index:
- name: roberta-base-squad2
  results: []
---

# roberta-base-squad2

This model is a fine-tuned version of [roberta-base](https://huggingface.co/roberta-base) on the squad_v2 dataset.

## Training and evaluation data

Trained and evaluated on the [squad_v2 dataset](https://huggingface.co/datasets/squad_v2).

## Training procedure

Trained on 16 Graphcore Mk2 IPUs using [optimum-graphcore](https://github.com/huggingface/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_v2 \
  --version_2_with_negative \
  --do_train \
  --do_eval \
  --num_train_epochs 3 \
  --per_device_train_batch_size 4 \
  --per_device_eval_batch_size 2 \
  --pod_type pod16 \
  --learning_rate 7e-5 \
  --max_seq_length 384 \
  --doc_stride 128 \
  --seed 1984 \
  --lr_scheduler_type linear \
  --loss_scaling 64 \
  --weight_decay 0.01 \
  --warmup_ratio 0.2 \
  --logging_steps 1 \
  --save_steps -1 \
  --dataloader_num_workers 64 \
  --output_dir roberta-base-squad2 \
  --overwrite_output_dir \
  --push_to_hub
```

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 7e-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: 0.2
- num_epochs: 3.0
- training precision: Mixed Precision

### Training results

```
***** train metrics *****
  epoch                    =        3.0
  train_loss               =     0.9982
  train_runtime            = 0:04:44.21
  train_samples            =     131823
  train_samples_per_second =    1391.43
  train_steps_per_second   =      5.425

***** eval metrics *****
  epoch                  =     3.0
  eval_HasAns_exact      = 78.1208
  eval_HasAns_f1         = 84.6569
  eval_HasAns_total      =    5928
  eval_NoAns_exact       = 82.0353
  eval_NoAns_f1          = 82.0353
  eval_NoAns_total       =    5945
  eval_best_exact        = 80.0809
  eval_best_exact_thresh =     0.0
  eval_best_f1           = 83.3442
  eval_best_f1_thresh    =     0.0
  eval_exact             = 80.0809
  eval_f1                = 83.3442
  eval_samples           =   12165
  eval_total             =   11873
```

### Framework versions

- Transformers 4.18.0.dev0
- Pytorch 1.10.0+cpu
- Datasets 2.0.0
- Tokenizers 0.11.6