File size: 3,641 Bytes
90f60b4
 
456945f
90f60b4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
tags:
- generation
language:
- multilingual
- cs
- en
---

# Mt5-base for Czech+English Generative Question Answering

This is the [mt5-base](https://huggingface.co/google/mt5-base) model with an LM head for a generation of extractive answers. In contrary to our [mt5-base-priming](https://huggingface.co/gaussalgo/mt5-base-priming-QA_en-cs/edit/main/README.md), this is a traditional sequence2sequence model without priming, though can also be used on other Text extraction tasks, such as Named Entity Recognition in zero-shot settings (with a significant decay in quality, compared to priming).

## Intended uses & limitations

This model is purposed to *generate* a segment of a given context that contains an answer to a given question (Extractive Question Answering) in English and Czech.
Given the fine-tuning on two languages and a good reported zero-shot cross-lingual applicability of other fine-tuned multilingual large language models, the model will likely also work on other languages as well, with a specific decay in quality.

Note that despite its size, English SQuAD has a variety of reported biases, 
conditioned by the relative position or type of the answer in the context that can affect the model's performance on new data 
(see, e.g. [L. Mikula (2022)](https://is.muni.cz/th/adh58/?lang=en), Chap. 4.1).

## Usage

Here is how to use this model to answer the question on a given context using 🤗 Transformers in PyTorch:

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("gaussalgo/mt5-base-generative-QA_en-cs")
model = AutoModelForSeq2SeqLM.from_pretrained("gaussalgo/mt5-base-generative-QA_en-cs")

context = """
Podle slovenského lidového podání byl Juro Jánošík obdařen magickými předměty (kouzelná valaška, čarovný opasek),
které mu dodávaly nadpřirozené schopnosti. Okrádal především šlechtice,
trestal panské dráby a ze svého lupu vyděloval část pro chudé, tedy bohatým bral a chudým dával.
"""
question = "Jaké schopnosti daly magické předměty Juro Jánošíkovi?"

inputs = tokenizer(question, context, return_tensors="pt")
outputs = model.generate(**inputs)

print("Answer:")
print(tokenizer.decode(outputs))

```

## Training

The model has been trained using [Adaptor library](https://github.com/gaussalgo/adaptor) v0.1.5, in parallel on both Czech and English data, with the following parameters:

```python
training_arguments = AdaptationArguments(output_dir="train_dir",
                                         learning_rate=5e-5,
                                         stopping_strategy=StoppingStrategy.ALL_OBJECTIVES_CONVERGED,
                                         do_train=True,
                                         do_eval=True,
                                         warmup_steps=1000,
                                         max_steps=100000,
                                         gradient_accumulation_steps=4,
                                         eval_steps=100,
                                         logging_steps=10,
                                         save_steps=1000,
                                         num_train_epochs=50,
                                         evaluation_strategy="steps",
                                         remove_unused_columns=False)

```

You can find the full training script in [train_mt5_qa_en+cs.py](https://huggingface.co/gaussalgo/mt5-base-generative-QA_en-cs/blob/main/train_mt5_qa_en%2Bcs.py), reproducible after a specific data preprocessing for Czech SQAD in [parse_czech_squad.py](parse_czech_squad.py)