Update README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
datasets:
|
| 3 |
+
- squad
|
| 4 |
+
tags:
|
| 5 |
+
- question-generation
|
| 6 |
+
- distilt5
|
| 7 |
+
- distilt5-qg
|
| 8 |
+
widget:
|
| 9 |
+
- text: "<hl> 42 <hl> is the answer to life, the universe and everything. </s>"
|
| 10 |
+
- text: "Python is a programming language. It is developed by <hl> Guido Van Rossum <hl>. </s>"
|
| 11 |
+
- text: "Although <hl> practicality <hl> beats purity </s>"
|
| 12 |
+
license: "MIT"
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
## DistilT5 for question-generation
|
| 16 |
+
This is distilled version of [t5-small-qa-qg-hl](https://huggingface.co/valhalla/t5-small-qa-qg-hl) model trained for answer aware question generation task. The answer spans are highlighted within the text with special highlight tokens.
|
| 17 |
+
|
| 18 |
+
The model is distilled using the **No Teacher Distillation** method proposed by Huggingface, [here](https://github.com/huggingface/transformers/tree/master/examples/seq2seq#distilbart).
|
| 19 |
+
|
| 20 |
+
We just copy alternating layers from `t5-small-qa-qg-hl` and finetune more on the same data. Following table lists other distilled models and their metrics.
|
| 21 |
+
|
| 22 |
+
| Name | BLEU-4 | METEOR | ROUGE-L | QA-EM | QA-F1 |
|
| 23 |
+
|---------------------------------------------------------------------------------|---------|---------|---------|--------|--------|
|
| 24 |
+
| [distilt5-qg-hl-6-4](https://huggingface.co/valhalla/distilt5-qg-hl-6-4) | 18.4141 | 24.8417 | 40.3435 | - | - |
|
| 25 |
+
| [distilt5-qa-qg-hl-6-4](https://huggingface.co/valhalla/distilt5-qa-qg-hl-6-4) | 18.6493 | 24.9685 | 40.5605 | 76.13 | 84.659 |
|
| 26 |
+
| [distilt5-qg-hl-12-6](https://huggingface.co/valhalla/distilt5-qg-hl-12-6) | 20.5275 | 26.5010 | 43.2676 | - | - |
|
| 27 |
+
| [distilt5-qa-qg-hl-12-6](https://huggingface.co/valhalla/distilt5-qa-qg-hl-12-6)| 20.6109 | 26.4533 | 43.0895 | 81.61 | 89.831 |
|
| 28 |
+
|
| 29 |
+
You can play with the model using the inference API, just highlight the answer spans with `<hl>` tokens. For example
|
| 30 |
+
|
| 31 |
+
`<hl> 42 <hl> is the answer to life, the universe and everything.`
|
| 32 |
+
|
| 33 |
+
For more deatils see [this](https://github.com/patil-suraj/question_generation) repo.
|
| 34 |
+
|
| 35 |
+
### Model in action 🚀
|
| 36 |
+
|
| 37 |
+
You'll need to clone the [repo](https://github.com/patil-suraj/question_generation).
|
| 38 |
+
|
| 39 |
+
[](https://colab.research.google.com/github/patil-suraj/question_generation/blob/master/question_generation.ipynb)
|
| 40 |
+
|
| 41 |
+
```python3
|
| 42 |
+
from pipelines import pipeline
|
| 43 |
+
nlp = pipeline("question-generation", model="valhalla/distilt5-qg-hl-6-4")
|
| 44 |
+
nlp("42 is the answer to life, universe and everything.")
|
| 45 |
+
=> [{'answer': '42', 'question': 'What is the answer to life?'}]
|
| 46 |
+
```
|