File size: 4,937 Bytes
ab63cfa |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 |
---
language: en
thumbnail:
license: mit
tags:
- question-answering
-
-
datasets:
- squad_v2
metrics:
- squad_v2
widget:
- text: "Where is the Eiffel Tower located?"
context: "The Eiffel Tower is a wrought-iron lattice tower on the Champ de Mars in Paris, France. It is named after the engineer Gustave Eiffel, whose company designed and built the tower."
- text: "Who is Frederic Chopin?"
context: "Frédéric François Chopin, born Fryderyk Franciszek Chopin (1 March 1810 – 17 October 1849), was a Polish composer and virtuoso pianist of the Romantic era who wrote primarily for solo piano."
---
## BERT-base uncased model fine-tuned on SQuAD v1
This model was created using the [nn_pruning](https://github.com/huggingface/nn_pruning) python library: the **linear layers contains 16.0%** of the original weights.
The model contains **24.0%** of the original weights **overall** (the embeddings account for a significant part of the model, and they are not pruned by this method).
With a simple resizing of the linear matrices it ran **2.63x as fast as bert-large-uncased-whole-word-masking** on the evaluation.
This is possible because the pruning method lead to structured matrices: to visualize them, hover below on the plot to see the non-zero/zero parts of each matrix.
<div class="graph"><script src="/madlag/bert-large-uncased-wwm-squadv2-x2.63-f82.6-d16-hybrid-v1/raw/main/model_card/density_info.js" id="cddd6c5c-2e1d-40c7-b172-f7d5422349a6"></script></div>
In terms of accuracy, its **F1 is 82.57**, compared with 85.85 for , a **F1 drop of 3.28**.
## Fine-Pruning details
This model was fine-tuned from the HuggingFace [model](https://huggingface.co/bert-large-uncased-whole-word-masking) uncased checkpoint on [SQuAD1.1](https://rajpurkar.github.io/SQuAD-explorer), and distilled from the model [madlag/bert-large-uncased-whole-word-masking-finetuned-squadv2](https://huggingface.co/madlag/bert-large-uncased-whole-word-masking-finetuned-squadv2).
This model is case-insensitive: it does not make a difference between english and English.
A side-effect of the block pruning is that some of the attention heads are completely removed: 190 heads were removed on a total of 384 (49.5%).
Here is a detailed view on how the remaining heads are distributed in the network after pruning.
<div class="graph"><script src="/madlag/bert-large-uncased-wwm-squadv2-x2.63-f82.6-d16-hybrid-v1/raw/main/model_card/pruning_info.js" id="03ad75cf-8048-44ae-a1d6-db69021cc168"></script></div>
## Details of the SQuAD1.1 dataset
| Dataset | Split | # samples |
| -------- | ----- | --------- |
| SQuAD 2.0 | train | 130.0K |
| SQuAD 2.0 | eval | 11.9k |
### Fine-tuning
- Python: `3.8.5`
- Machine specs:
```CPU: Intel(R) Core(TM) i7-6700K CPU
Memory: 64 GiB
GPUs: 1 GeForce GTX 3090, with 24GiB memory
GPU driver: 455.23.05, CUDA: 11.1
```
### Results
**Pytorch model file size**: `1084MB` (original BERT: `1228.0MB`)
| Metric | # Value | # Original ([Table 2](https://www.aclweb.org/anthology/N19-1423.pdf))| Variation |
| ------ | --------- | --------- | --------- |
| **EM** | **79.70** | **82.83** | **-4.13**|
| **F1** | **82.57** | **85.85** | **-3.28**|
```
{
"HasAns_exact": 74.8144399460189,
"HasAns_f1": 80.555306012496,
"HasAns_total": 5928,
"NoAns_exact": 84.57527333894029,
"NoAns_f1": 84.57527333894029,
"NoAns_total": 5945,
"best_exact": 79.70184452118251,
"best_exact_thresh": 0.0,
"best_f1": 82.56816761071966,
"best_f1_thresh": 0.0,
"exact": 79.70184452118251,
"f1": 82.56816761071981,
"total": 11873
}
```
## Example Usage
Install nn_pruning: it contains the optimization script, which just pack the linear layers into smaller ones by removing empty rows/columns.
`pip install nn_pruning`
Then you can use the `transformers library` almost as usual: you just have to call `optimize_model` when the pipeline has loaded.
```python
from transformers import pipeline
from nn_pruning.inference_model_patcher import optimize_model
qa_pipeline = pipeline(
"question-answering",
model="madlag/bert-large-uncased-wwm-squadv2-x2.63-f82.6-d16-hybrid-v1",
tokenizer="madlag/bert-large-uncased-wwm-squadv2-x2.63-f82.6-d16-hybrid-v1"
)
print("bert-large-uncased-whole-word-masking parameters: 445.0M")
print(f"Parameters count (includes only head pruning, not feed forward pruning)={int(qa_pipeline.model.num_parameters() / 1E6)}M")
qa_pipeline.model = optimize_model(qa_pipeline.model, "dense")
print(f"Parameters count after complete optimization={int(qa_pipeline.model.num_parameters() / 1E6)}M")
predictions = qa_pipeline({
'context': "Frédéric François Chopin, born Fryderyk Franciszek Chopin (1 March 1810 – 17 October 1849), was a Polish composer and virtuoso pianist of the Romantic era who wrote primarily for solo piano.",
'question': "Who is Frederic Chopin?",
})
print("Predictions", predictions)
``` |