If you're opening this Notebook on colab, you will probably need to install 🤗 Transformers and 🤗 Datasets. Uncomment the following cell and run it.
#! pip install datasets transformers
If you're opening this notebook locally, make sure your environment has an install from the last version of those libraries.
To be able to share your model with the community and generate results like the one shown in the picture below via the inference API, there are a few more steps to follow.
First you have to store your authentication token from the Hugging Face website (sign up here if you haven't already!) then execute the following cell and input your username and password:
from huggingface_hub import notebook_login
notebook_login()
Then you need to install Git-LFS. Uncomment the following instructions:
# !apt install git-lfs
Make sure your version of Transformers is at least 4.11.0 since the functionality was introduced in that version:
import transformers
print(transformers.__version__)
You can find a script version of this notebook to fine-tune your model in a distributed fashion using multiple GPUs or TPUs here.
We also quickly upload some telemetry - this tells us which examples and software versions are getting used so we know where to prioritize our maintenance efforts. We don't collect (or care about) any personally identifiable information, but if you'd prefer not to be counted, feel free to skip this step or delete this cell entirely.
from transformers.utils import send_example_telemetry
send_example_telemetry("question_answering_notebook", framework="pytorch")
Fine-tuning a model on a question-answering task¶
In this notebook, we will see how to fine-tune one of the 🤗 Transformers model to a question answering task, which is the task of extracting the answer to a question from a given context. We will see how to easily load a dataset for these kinds of tasks and use the Trainer API to fine-tune a model on it.
Note: This notebook finetunes models that answer question by taking a substring of a context, not by generating new text.
This notebook is built to run on any question answering task with the same format as SQUAD (version 1 or 2), with any model checkpoint from the Model Hub as long as that model has a version with a token classification head and a fast tokenizer (check on this table if this is the case). It might just need some small adjustments if you decide to use a different dataset than the one used here. Depending on you model and the GPU you are using, you might need to adjust the batch size to avoid out-of-memory errors. Set those three parameters, then the rest of the notebook should run smoothly:
# This flag is the difference between SQUAD v1 or 2 (if you're using another dataset, it indicates if impossible
# answers are allowed or not).
squad_v2 = False
model_checkpoint = "distilbert-base-uncased"
batch_size = 16
Loading the dataset¶
We will use the 🤗 Datasets library to download the data and get the metric we need to use for evaluation (to compare our model to the benchmark). This can be easily done with the functions load_dataset and load_metric.
from datasets import load_dataset, load_metric
For our example here, we'll use the SQUAD dataset. The notebook should work with any question answering dataset provided by the 🤗 Datasets library. If you're using your own dataset defined from a JSON or csv file (see the Datasets documentation on how to load them), it might need some adjustments in the names of the columns used.
datasets = load_dataset("squad_v2" if squad_v2 else "squad")
Reusing dataset squad (/home/sgugger/.cache/huggingface/datasets/squad/plain_text/1.0.0/4c81550d83a2ac7c7ce23783bd8ff36642800e6633c1f18417fb58c3ff50cdd7)
The datasets object itself is DatasetDict, which contains one key for the training, validation and test set.
datasets
DatasetDict({
train: Dataset({
features: ['id', 'title', 'context', 'question', 'answers'],
num_rows: 87599
})
validation: Dataset({
features: ['id', 'title', 'context', 'question', 'answers'],
num_rows: 10570
})
})
We can see the training, validation and test sets all have a column for the context, the question and the answers to those questions.
To access an actual element, you need to select a split first, then give an index:
datasets["train"][0]
{'answers': {'answer_start': [515], 'text': ['Saint Bernadette Soubirous']},
'context': 'Architecturally, the school has a Catholic character. Atop the Main Building\'s gold dome is a golden statue of the Virgin Mary. Immediately in front of the Main Building and facing it, is a copper statue of Christ with arms upraised with the legend "Venite Ad Me Omnes". Next to the Main Building is the Basilica of the Sacred Heart. Immediately behind the basilica is the Grotto, a Marian place of prayer and reflection. It is a replica of the grotto at Lourdes, France where the Virgin Mary reputedly appeared to Saint Bernadette Soubirous in 1858. At the end of the main drive (and in a direct line that connects through 3 statues and the Gold Dome), is a simple, modern stone statue of Mary.',
'id': '5733be284776f41900661182',
'question': 'To whom did the Virgin Mary allegedly appear in 1858 in Lourdes France?',
'title': 'University_of_Notre_Dame'}
We can see the answers are indicated by their start position in the text (here at character 515) and their full text, which is a substring of the context as we mentioned above.
To get a sense of what the data looks like, the following function will show some examples picked randomly in the dataset (automatically decoding the labels in passing).
from datasets import ClassLabel, Sequence
import random
import pandas as pd
from IPython.display import display, HTML
def show_random_elements(dataset, num_examples=10):
assert num_examples <= len(dataset), "Can't pick more elements than there are in the dataset."
picks = []
for _ in range(num_examples):
pick = random.randint(0, len(dataset)-1)
while pick in picks:
pick = random.randint(0, len(dataset)-1)
picks.append(pick)
df = pd.DataFrame(dataset[picks])
for column, typ in dataset.features.items():
if isinstance(typ, ClassLabel):
df[column] = df[column].transform(lambda i: typ.names[i])
elif isinstance(typ, Sequence) and isinstance(typ.feature, ClassLabel):
df[column] = df[column].transform(lambda x: [typ.feature.names[i] for i in x])
display(HTML(df.to_html()))
show_random_elements(datasets["train"])
| answers | context | id | question | title | |
|---|---|---|---|---|---|
| 0 | {'answer_start': [595], 'text': ['1964']} | Paul VI opened the third period on 14 September 1964, telling the Council Fathers that he viewed the text about the Church as the most important document to come out from the Council. As the Council discussed the role of bishops in the papacy, Paul VI issued an explanatory note confirming the primacy of the papacy, a step which was viewed by some as meddling in the affairs of the Council American bishops pushed for a speedy resolution on religious freedom, but Paul VI insisted this to be approved together with related texts such as ecumenism. The Pope concluded the session on 21 November 1964, with the formal pronouncement of Mary as Mother of the Church. | 5726bc075951b619008f7c63 | In what year did Paul VI formally appoint Mary as mother of the Catholic church? | Pope_Paul_VI |
| 1 | {'answer_start': [453], 'text': ['asiento']} | At the concluding Treaty of Utrecht, Philip renounced his and his descendants' right to the French throne and Spain lost its empire in Europe. The British Empire was territorially enlarged: from France, Britain gained Newfoundland and Acadia, and from Spain, Gibraltar and Minorca. Gibraltar became a critical naval base and allowed Britain to control the Atlantic entry and exit point to the Mediterranean. Spain also ceded the rights to the lucrative asiento (permission to sell slaves in Spanish America) to Britain. | 57266a9bf1498d1400e8df18 | What was the Spanish term for permission to sell slaves in Spanish America? | British_Empire |
| 2 | {'answer_start': [0], 'text': ['The variances in nomenclature']} | The variances in nomenclature in the region spanned by the Alps makes classification of the mountains and subregions difficult, but a general classification is that of the Eastern Alps and Western Alps with the divide between the two occurring in eastern Switzerland according to geologist Stefan Schmid, near the Splügen Pass. | 56f876bda6d7ea1400e17689 | What makes the classification of the mountains and subregions difficult? | Alps |
| 3 | {'answer_start': [727], 'text': ['animal electricity']} | William Champion's brother, John, patented a process in 1758 for calcining zinc sulfide into an oxide usable in the retort process. Prior to this, only calamine could be used to produce zinc. In 1798, Johann Christian Ruberg improved on the smelting process by building the first horizontal retort smelter. Jean-Jacques Daniel Dony built a different kind of horizontal zinc smelter in Belgium, which processed even more zinc. Italian doctor Luigi Galvani discovered in 1780 that connecting the spinal cord of a freshly dissected frog to an iron rail attached by a brass hook caused the frog's leg to twitch. He incorrectly thought he had discovered an ability of nerves and muscles to create electricity and called the effect "animal electricity". The galvanic cell and the process of galvanization were both named for Luigi Galvani and these discoveries paved the way for electrical batteries, galvanization and cathodic protection. | 572b5939f75d5e190021fd95 | What did Galvani name the effect he created of causing the frogs legs to twitch? | Zinc |
| 4 | {'answer_start': [150], 'text': ['the Jurchens']} | The Qing dynasty (1644–1911) was founded after the fall of the Ming, the last Han Chinese dynasty, by the Manchus. The Manchus were formerly known as the Jurchens. When Beijing was captured by Li Zicheng's peasant rebels in 1644, the Chongzhen Emperor, the last Ming emperor, committed suicide. The Manchus then allied with former Ming general Wu Sangui and seized control of Beijing, which became the new capital of the Qing dynasty. The Mancus adopted the Confucian norms of traditional Chinese government in their rule of China proper. Schoppa, the editor of The Columbia Guide to Modern Chinese History argues, "A date around 1780 as the beginning of modern China is thus closer to what we know today as historical 'reality'. It also allows us to have a better baseline to understand the precipitous decline of the Chinese polity in the nineteenth and twentieth centuries." | 572ee763cb0c0d14000f1666 | What were the Manchus originally known as? | Modern_history |
| 5 | {'answer_start': [145], 'text': ['song choice and performance']} | Beginning in the tenth season[citation needed], permanent mentors were brought in during the live shows to help guide the contestants with their song choice and performance. Jimmy Iovine was the mentor in the tenth through twelfth seasons, former judge Randy Jackson was the mentor for the thirteenth season and Scott Borchetta was the mentor for the fourteenth and fifteenth season. The mentors regularly bring in guest mentors to aid them, including Akon, Alicia Keys, Lady Gaga, and current judge Harry Connick, Jr.. | 56daf310e7c41114004b4b74 | What two things did the mentors help the contestants with? | American_Idol |
| 6 | {'answer_start': [191], 'text': ['Kamran Baghirov']} | Gorbachev refused to make any changes to the status of Nagorno Karabakh, which remained part of Azerbaijan. He instead sacked the Communist Party Leaders in both Republics – on May 21, 1988, Kamran Baghirov was replaced by Abdulrahman Vezirov as First Secretary of the Azerbaijan Communist Party. From July 23 to September 1988, a group of Azerbaijani intellectuals began working for a new organization called the Popular Front of Azerbaijan, loosely based on the Estonian Popular Front. On September 17, when gun battles broke out between the Armenians and Azerbaijanis near Stepanakert, two soldiers were killed and more than two dozen injured. This led to almost tit-for-tat ethnic polarization in Nagorno-Karabakh's two main towns: The Azerbaijani minority was expelled from Stepanakert, and the Armenian minority was expelled from Shusha. On November 17, 1988, in response to the exodus of tens of thousands of Azerbaijanis from Armenia, a series of mass demonstrations began in Baku's Lenin Square, lasting 18 days and attracting half a million demonstrators. On December 5, 1988, the Soviet militia finally moved in, cleared the square by force, and imposed a curfew that lasted ten months. | 57279998dd62a815002ea1a5 | Who was First Secretary prior to Vezirov? | Dissolution_of_the_Soviet_Union |
| 7 | {'answer_start': [0], 'text': ['World literature']} | World literature was enriched by the works of Edmund Spenser, John Milton, John Bunyan, John Donne, John Dryden, Daniel Defoe, William Wordsworth, Jonathan Swift, Johann Wolfgang Goethe, Friedrich Schiller, Samuel Taylor Coleridge, Edgar Allan Poe, Matthew Arnold, Conrad Ferdinand Meyer, Theodor Fontane, Washington Irving, Robert Browning, Emily Dickinson, Emily Brontë, Charles Dickens, Nathaniel Hawthorne, Thomas Stearns Eliot, John Galsworthy, Thomas Mann, William Faulkner, John Updike, and many others. | 57325d03b9d445190005eaac | Samuel Taylor is listed as enriching what? | Protestantism |
| 8 | {'answer_start': [85], 'text': ['1870']} | The first road connecting the city to the mainland at Pleasantville was completed in 1870 and charged a 30-cent toll. Albany Avenue was the first road to the mainland that was available without a toll. | 57060ece75f01819005e791b | The first road that connected Atlantic City to the mainland was completed in what year? | Atlantic_City,_New_Jersey |
| 9 | {'answer_start': [172], 'text': ['progressively eliminate child labour']} | In addition to setting the international law, the United Nations initiated International Program on the Elimination of Child Labour (IPEC) in 1992. This initiative aims to progressively eliminate child labour through strengthening national capacities to address some of the causes of child labour. Amongst the key initiative is the so-called time-bounded programme countries, where child labour is most prevalent and schooling opportunities lacking. The initiative seeks to achieve amongst other things, universal primary school availability. The IPEC has expanded to at least the following target countries: Bangladesh, Brazil, China, Egypt, India, Indonesia, Mexico, Nigeria, Pakistan, Democratic Republic of Congo, El Salvador, Nepal, Tanzania, Dominican Republic, Costa Rica, Philippines, Senegal, South Africa and Turkey. | 5727972a708984140094e1a8 | What is the aim of this? | Child_labour |
Preprocessing the training data¶
Before we can feed those texts to our model, we need to preprocess them. This is done by a 🤗 Transformers Tokenizer which will (as the name indicates) tokenize the inputs (including converting the tokens to their corresponding IDs in the pretrained vocabulary) and put it in a format the model expects, as well as generate the other inputs that model requires.
To do all of this, we instantiate our tokenizer with the AutoTokenizer.from_pretrained method, which will ensure:
- we get a tokenizer that corresponds to the model architecture we want to use,
- we download the vocabulary used when pretraining this specific checkpoint.
That vocabulary will be cached, so it's not downloaded again the next time we run the cell.
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
The following assertion ensures that our tokenizer is a fast tokenizers (backed by Rust) from the 🤗 Tokenizers library. Those fast tokenizers are available for almost all models, and we will need some of the special features they have for our preprocessing.
import transformers
assert isinstance(tokenizer, transformers.PreTrainedTokenizerFast)
You can check which type of models have a fast tokenizer available and which don't on the big table of models.
You can directly call this tokenizer on two sentences (one for the answer, one for the context):
tokenizer("What is your name?", "My name is Sylvain.")
{'input_ids': [101, 2054, 2003, 2115, 2171, 1029, 102, 2026, 2171, 2003, 25353, 22144, 2378, 1012, 102], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
Depending on the model you selected, you will see different keys in the dictionary returned by the cell above. They don't matter much for what we're doing here (just know they are required by the model we will instantiate later), you can learn more about them in this tutorial if you're interested.
Now one specific thing for the preprocessing in question answering is how to deal with very long documents. We usually truncate them in other tasks, when they are longer than the model maximum sentence length, but here, removing part of the the context might result in losing the answer we are looking for. To deal with this, we will allow one (long) example in our dataset to give several input features, each of length shorter than the maximum length of the model (or the one we set as a hyper-parameter). Also, just in case the answer lies at the point we split a long context, we allow some overlap between the features we generate controlled by the hyper-parameter doc_stride:
max_length = 384 # The maximum length of a feature (question and context)
doc_stride = 128 # The authorized overlap between two part of the context when splitting it is needed.
Let's find one long example in our dataset:
for i, example in enumerate(datasets["train"]):
if len(tokenizer(example["question"], example["context"])["input_ids"]) > 384:
break
example = datasets["train"][i]
Without any truncation, we get the following length for the input IDs:
len(tokenizer(example["question"], example["context"])["input_ids"])
396
Now, if we just truncate, we will lose information (and possibly the answer to our question):
len(tokenizer(example["question"], example["context"], max_length=max_length, truncation="only_second")["input_ids"])
384
Note that we never want to truncate the question, only the context, else the only_second truncation picked. Now, our tokenizer can automatically return us a list of features capped by a certain maximum length, with the overlap we talked above, we just have to tell it with return_overflowing_tokens=True and by passing the stride:
tokenized_example = tokenizer(
example["question"],
example["context"],
max_length=max_length,
truncation="only_second",
return_overflowing_tokens=True,
stride=doc_stride
)
Now we don't have one list of input_ids, but several:
[len(x) for x in tokenized_example["input_ids"]]
[384, 157]
And if we decode them, we can see the overlap:
for x in tokenized_example["input_ids"][:2]:
print(tokenizer.decode(x))
[CLS] how many wins does the notre dame men's basketball team have? [SEP] the men's basketball team has over 1, 600 wins, one of only 12 schools who have reached that mark, and have appeared in 28 ncaa tournaments. former player austin carr holds the record for most points scored in a single game of the tournament with 61. although the team has never won the ncaa tournament, they were named by the helms athletic foundation as national champions twice. the team has orchestrated a number of upsets of number one ranked teams, the most notable of which was ending ucla's record 88 - game winning streak in 1974. the team has beaten an additional eight number - one teams, and those nine wins rank second, to ucla's 10, all - time in wins against the top team. the team plays in newly renovated purcell pavilion ( within the edmund p. joyce center ), which reopened for the beginning of the 2009 – 2010 season. the team is coached by mike brey, who, as of the 2014 – 15 season, his fifteenth at notre dame, has achieved a 332 - 165 record. in 2009 they were invited to the nit, where they advanced to the semifinals but were beaten by penn state who went on and beat baylor in the championship. the 2010 – 11 team concluded its regular season ranked number seven in the country, with a record of 25 – 5, brey's fifth straight 20 - win season, and a second - place finish in the big east. during the 2014 - 15 season, the team went 32 - 6 and won the acc conference tournament, later advancing to the elite 8, where the fighting irish lost on a missed buzzer - beater against then undefeated kentucky. led by nba draft picks jerian grant and pat connaughton, the fighting irish beat the eventual national champion duke blue devils twice during the season. the 32 wins were [SEP] [CLS] how many wins does the notre dame men's basketball team have? [SEP] championship. the 2010 – 11 team concluded its regular season ranked number seven in the country, with a record of 25 – 5, brey's fifth straight 20 - win season, and a second - place finish in the big east. during the 2014 - 15 season, the team went 32 - 6 and won the acc conference tournament, later advancing to the elite 8, where the fighting irish lost on a missed buzzer - beater against then undefeated kentucky. led by nba draft picks jerian grant and pat connaughton, the fighting irish beat the eventual national champion duke blue devils twice during the season. the 32 wins were the most by the fighting irish team since 1908 - 09. [SEP]
Now this will give us some work to properly treat the answers: we need to find in which of those features the answer actually is, and where exactly in that feature. The models we will use require the start and end positions of these answers in the tokens, so we will also need to to map parts of the original context to some tokens. Thankfully, the tokenizer we're using can help us with that by returning an offset_mapping:
tokenized_example = tokenizer(
example["question"],
example["context"],
max_length=max_length,
truncation="only_second",
return_overflowing_tokens=True,
return_offsets_mapping=True,
stride=doc_stride
)
print(tokenized_example["offset_mapping"][0][:100])
[(0, 0), (0, 3), (4, 8), (9, 13), (14, 18), (19, 22), (23, 28), (29, 33), (34, 37), (37, 38), (38, 39), (40, 50), (51, 55), (56, 60), (60, 61), (0, 0), (0, 3), (4, 7), (7, 8), (8, 9), (10, 20), (21, 25), (26, 29), (30, 34), (35, 36), (36, 37), (37, 40), (41, 45), (45, 46), (47, 50), (51, 53), (54, 58), (59, 61), (62, 69), (70, 73), (74, 78), (79, 86), (87, 91), (92, 96), (96, 97), (98, 101), (102, 106), (107, 115), (116, 118), (119, 121), (122, 126), (127, 138), (138, 139), (140, 146), (147, 153), (154, 160), (161, 165), (166, 171), (172, 175), (176, 182), (183, 186), (187, 191), (192, 198), (199, 205), (206, 208), (209, 210), (211, 217), (218, 222), (223, 225), (226, 229), (230, 240), (241, 245), (246, 248), (248, 249), (250, 258), (259, 262), (263, 267), (268, 271), (272, 277), (278, 281), (282, 285), (286, 290), (291, 301), (301, 302), (303, 307), (308, 312), (313, 318), (319, 321), (322, 325), (326, 330), (330, 331), (332, 340), (341, 351), (352, 354), (355, 363), (364, 373), (374, 379), (379, 380), (381, 384), (385, 389), (390, 393), (394, 406), (407, 408), (409, 415), (416, 418)]
This gives, for each index of our input IDS, the corresponding start and end character in the original text that gave our token. The very first token ([CLS]) has (0, 0) because it doesn't correspond to any part of the question/answer, then the second token is the same as the characters 0 to 3 of the question:
first_token_id = tokenized_example["input_ids"][0][1]
offsets = tokenized_example["offset_mapping"][0][1]
print(tokenizer.convert_ids_to_tokens([first_token_id])[0], example["question"][offsets[0]:offsets[1]])
how How
So we can use this mapping to find the position of the start and end tokens of our answer in a given feature. We just have to distinguish which parts of the offsets correspond to the question and which part correspond to the context, this is where the sequence_ids method of our tokenized_example can be useful:
sequence_ids = tokenized_example.sequence_ids()
print(sequence_ids)
[None, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, None, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, None]
It returns None for the special tokens, then 0 or 1 depending on whether the corresponding token comes from the first sentence past (the question) or the second (the context). Now with all of this, we can find the first and last token of the answer in one of our input feature (or if the answer is not in this feature):
answers = example["answers"]
start_char = answers["answer_start"][0]
end_char = start_char + len(answers["text"][0])
# Start token index of the current span in the text.
token_start_index = 0
while sequence_ids[token_start_index] != 1:
token_start_index += 1
# End token index of the current span in the text.
token_end_index = len(tokenized_example["input_ids"][0]) - 1
while sequence_ids[token_end_index] != 1:
token_end_index -= 1
# Detect if the answer is out of the span (in which case this feature is labeled with the CLS index).
offsets = tokenized_example["offset_mapping"][0]
if (offsets[token_start_index][0] <= start_char and offsets[token_end_index][1] >= end_char):
# Move the token_start_index and token_end_index to the two ends of the answer.
# Note: we could go after the last offset if the answer is the last word (edge case).
while token_start_index < len(offsets) and offsets[token_start_index][0] <= start_char:
token_start_index += 1
start_position = token_start_index - 1
while offsets[token_end_index][1] >= end_char:
token_end_index -= 1
end_position = token_end_index + 1
print(start_position, end_position)
else:
print("The answer is not in this feature.")
23 26
And we can double check that it is indeed the theoretical answer:
print(tokenizer.decode(tokenized_example["input_ids"][0][start_position: end_position+1]))
print(answers["text"][0])
over 1, 600 over 1,600
For this notebook to work with any kind of models, we need to account for the special case where the model expects padding on the left (in which case we switch the order of the question and the context):
pad_on_right = tokenizer.padding_side == "right"
Now let's put everything together in one function we will apply to our training set. In the case of impossible answers (the answer is in another feature given by an example with a long context), we set the cls index for both the start and end position. We could also simply discard those examples from the training set if the flag allow_impossible_answers is False. Since the preprocessing is already complex enough as it is, we've kept is simple for this part.
def prepare_train_features(examples):
# Some of the questions have lots of whitespace on the left, which is not useful and will make the
# truncation of the context fail (the tokenized question will take a lots of space). So we remove that
# left whitespace
examples["question"] = [q.lstrip() for q in examples["question"]]
# Tokenize our examples with truncation and padding, but keep the overflows using a stride. This results
# in one example possible giving several features when a context is long, each of those features having a
# context that overlaps a bit the context of the previous feature.
tokenized_examples = tokenizer(
examples["question" if pad_on_right else "context"],
examples["context" if pad_on_right else "question"],
truncation="only_second" if pad_on_right else "only_first",
max_length=max_length,
stride=doc_stride,
return_overflowing_tokens=True,
return_offsets_mapping=True,
padding="max_length",
)
# Since one example might give us several features if it has a long context, we need a map from a feature to
# its corresponding example. This key gives us just that.
sample_mapping = tokenized_examples.pop("overflow_to_sample_mapping")
# The offset mappings will give us a map from token to character position in the original context. This will
# help us compute the start_positions and end_positions.
offset_mapping = tokenized_examples.pop("offset_mapping")
# Let's label those examples!
tokenized_examples["start_positions"] = []
tokenized_examples["end_positions"] = []
for i, offsets in enumerate(offset_mapping):
# We will label impossible answers with the index of the CLS token.
input_ids = tokenized_examples["input_ids"][i]
cls_index = input_ids.index(tokenizer.cls_token_id)
# Grab the sequence corresponding to that example (to know what is the context and what is the question).
sequence_ids = tokenized_examples.sequence_ids(i)
# One example can give several spans, this is the index of the example containing this span of text.
sample_index = sample_mapping[i]
answers = examples["answers"][sample_index]
# If no answers are given, set the cls_index as answer.
if len(answers["answer_start"]) == 0:
tokenized_examples["start_positions"].append(cls_index)
tokenized_examples["end_positions"].append(cls_index)
else:
# Start/end character index of the answer in the text.
start_char = answers["answer_start"][0]
end_char = start_char + len(answers["text"][0])
# Start token index of the current span in the text.
token_start_index = 0
while sequence_ids[token_start_index] != (1 if pad_on_right else 0):
token_start_index += 1
# End token index of the current span in the text.
token_end_index = len(input_ids) - 1
while sequence_ids[token_end_index] != (1 if pad_on_right else 0):
token_end_index -= 1
# Detect if the answer is out of the span (in which case this feature is labeled with the CLS index).
if not (offsets[token_start_index][0] <= start_char and offsets[token_end_index][1] >= end_char):
tokenized_examples["start_positions"].append(cls_index)
tokenized_examples["end_positions"].append(cls_index)
else:
# Otherwise move the token_start_index and token_end_index to the two ends of the answer.
# Note: we could go after the last offset if the answer is the last word (edge case).
while token_start_index < len(offsets) and offsets[token_start_index][0] <= start_char:
token_start_index += 1
tokenized_examples["start_positions"].append(token_start_index - 1)
while offsets[token_end_index][1] >= end_char:
token_end_index -= 1
tokenized_examples["end_positions"].append(token_end_index + 1)
return tokenized_examples
This function works with one or several examples. In the case of several examples, the tokenizer will return a list of lists for each key:
features = prepare_train_features(datasets['train'][:5])
To apply this function on all the sentences (or pairs of sentences) in our dataset, we just use the map method of our dataset object we created earlier. This will apply the function on all the elements of all the splits in dataset, so our training, validation and testing data will be preprocessed in one single command. Since our preprocessing changes the number of samples, we need to remove the old columns when applying it.
tokenized_datasets = datasets.map(prepare_train_features, batched=True, remove_columns=datasets["train"].column_names)
Loading cached processed dataset at /home/sgugger/.cache/huggingface/datasets/squad/plain_text/1.0.0/4c81550d83a2ac7c7ce23783bd8ff36642800e6633c1f18417fb58c3ff50cdd7/cache-a5c71e98733887b0.arrow Loading cached processed dataset at /home/sgugger/.cache/huggingface/datasets/squad/plain_text/1.0.0/4c81550d83a2ac7c7ce23783bd8ff36642800e6633c1f18417fb58c3ff50cdd7/cache-14932a8c6aecc96d.arrow
Even better, the results are automatically cached by the 🤗 Datasets library to avoid spending time on this step the next time you run your notebook. The 🤗 Datasets library is normally smart enough to detect when the function you pass to map has changed (and thus requires to not use the cache data). For instance, it will properly detect if you change the task in the first cell and rerun the notebook. 🤗 Datasets warns you when it uses cached files, you can pass load_from_cache_file=False in the call to map to not use the cached files and force the preprocessing to be applied again.
Note that we passed batched=True to encode the texts by batches together. This is to leverage the full benefit of the fast tokenizer we loaded earlier, which will use multi-threading to treat the texts in a batch concurrently.
Fine-tuning the model¶
Now that our data is ready for training, we can download the pretrained model and fine-tune it. Since our task is question answering, we use the AutoModelForQuestionAnswering class. Like with the tokenizer, the from_pretrained method will download and cache the model for us:
from transformers import AutoModelForQuestionAnswering, TrainingArguments, Trainer
model = AutoModelForQuestionAnswering.from_pretrained(model_checkpoint)
Some weights of the model checkpoint at distilbert-base-uncased were not used when initializing DistilBertForQuestionAnswering: ['vocab_transform.weight', 'vocab_transform.bias', 'vocab_layer_norm.weight', 'vocab_layer_norm.bias', 'vocab_projector.weight', 'vocab_projector.bias'] - This IS expected if you are initializing DistilBertForQuestionAnswering from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model). - This IS NOT expected if you are initializing DistilBertForQuestionAnswering from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model). Some weights of DistilBertForQuestionAnswering were not initialized from the model checkpoint at distilbert-base-uncased and are newly initialized: ['qa_outputs.weight', 'qa_outputs.bias'] You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
The warning is telling us we are throwing away some weights (the vocab_transform and vocab_layer_norm layers) and randomly initializing some other (the pre_classifier and classifier layers). This is absolutely normal in this case, because we are removing the head used to pretrain the model on a masked language modeling objective and replacing it with a new head for which we don't have pretrained weights, so the library warns us we should fine-tune this model before using it for inference, which is exactly what we are going to do.
To instantiate a Trainer, we will need to define three more things. The most important is the TrainingArguments, which is a class that contains all the attributes to customize the training. It requires one folder name, which will be used to save the checkpoints of the model, and all other arguments are optional:
model_name = model_checkpoint.split("/")[-1]
args = TrainingArguments(
f"{model_name}-finetuned-squad",
evaluation_strategy = "epoch",
learning_rate=2e-5,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size,
num_train_epochs=3,
weight_decay=0.01,
push_to_hub=True,
)
Here we set the evaluation to be done at the end of each epoch, tweak the learning rate, use the batch_size defined at the top of the notebook and customize the number of epochs for training, as well as the weight decay.
The last argument to setup everything so we can push the model to the Hub regularly during training. Remove it if you didn't follow the installation steps at the top of the notebook. If you want to save your model locally in a name that is different than the name of the repository it will be pushed, or if you want to push your model under an organization and not your name space, use the hub_model_id argument to set the repo name (it needs to be the full name, including your namespace: for instance "sgugger/bert-finetuned-squad" or "huggingface/bert-finetuned-squad").
Then we will need a data collator that will batch our processed examples together, here the default one will work:
from transformers import default_data_collator
data_collator = default_data_collator
We will evaluate our model and compute metrics in the next section (this is a very long operation, so we will only compute the evaluation loss during training).
Then we just need to pass all of this along with our datasets to the Trainer:
trainer = Trainer(
model,
args,
train_dataset=tokenized_datasets["train"],
eval_dataset=tokenized_datasets["validation"],
data_collator=data_collator,
tokenizer=tokenizer,
)
We can now finetune our model by just calling the train method:
trainer.train()
| Epoch | Training Loss | Validation Loss | Runtime | Samples Per Second |
|---|---|---|---|---|
| 1 | 1.220600 | 1.160322 | 39.574900 | 272.496000 |
| 2 | 0.945200 | 1.121690 | 39.706000 | 271.596000 |
| 3 | 0.773000 | 1.157358 | 39.734000 | 271.405000 |
TrainOutput(global_step=16599, training_loss=1.1112074395519933, metrics={'train_runtime': 3487.0114, 'train_samples_per_second': 4.76, 'total_flos': 40606919924189184, 'epoch': 3.0})
Since this training is particularly long, let's save the model just in case we need to restart.
trainer.save_model("test-squad-trained")
Evaluation¶
Evaluating our model will require a bit more work, as we will need to map the predictions of our model back to parts of the context. The model itself predicts logits for the start and en position of our answers: if we take a batch from our validation datalaoder, here is the output our model gives us:
import torch
for batch in trainer.get_eval_dataloader():
break
batch = {k: v.to(trainer.args.device) for k, v in batch.items()}
with torch.no_grad():
output = trainer.model(**batch)
output.keys()
odict_keys(['loss', 'start_logits', 'end_logits'])
The output of the model is a dict-like object that contains the loss (since we provided labels), the start and end logits. We won't need the loss for our predictions, let's have a look a the logits:
output.start_logits.shape, output.end_logits.shape
(torch.Size([16, 384]), torch.Size([16, 384]))
We have one logit for each feature and each token. The most obvious thing to predict an answer for each featyre is to take the index for the maximum of the start logits as a start position and the index of the maximum of the end logits as an end position.
output.start_logits.argmax(dim=-1), output.end_logits.argmax(dim=-1)
(tensor([ 46, 57, 78, 43, 118, 15, 72, 35, 15, 34, 73, 41, 80, 91,
156, 35], device='cuda:0'),
tensor([ 47, 58, 81, 55, 118, 110, 75, 37, 110, 36, 76, 53, 83, 94,
158, 35], device='cuda:0'))
This will work great in a lot of cases, but what if this prediction gives us something impossible: the start position could be greater than the end position, or point to a span of text in the question instead of the answer. In that case, we might want to look at the second best prediction to see if it gives a possible answer and select that instead.
However, picking the second best answer is not as easy as picking the best one: is it the second best index in the start logits with the best index in the end logits? Or the best index in the start logits with the second best index in the end logits? And if that second best answer is not possible either, it gets even trickier for the third best answer.
To classify our answers, we will use the score obtained by adding the start and end logits. We won't try to order all the possible answers and limit ourselves to with a hyper-parameter we call n_best_size. We'll pick the best indices in the start and end logits and gather all the answers this predicts. After checking if each one is valid, we will sort them by their score and keep the best one. Here is how we would do this on the first feature in the batch:
n_best_size = 20
import numpy as np
start_logits = output.start_logits[0].cpu().numpy()
end_logits = output.end_logits[0].cpu().numpy()
# Gather the indices the best start/end logits:
start_indexes = np.argsort(start_logits)[-1 : -n_best_size - 1 : -1].tolist()
end_indexes = np.argsort(end_logits)[-1 : -n_best_size - 1 : -1].tolist()
valid_answers = []
for start_index in start_indexes:
for end_index in end_indexes:
if start_index <= end_index: # We need to refine that test to check the answer is inside the context
valid_answers.append(
{
"score": start_logits[start_index] + end_logits[end_index],
"text": "" # We need to find a way to get back the original substring corresponding to the answer in the context
}
)
And then we can sort the valid_answers according to their score and only keep the best one. The only point left is how to check a given span is inside the context (and not the question) and how to get back the text inside. To do this, we need to add two things to our validation features:
- the ID of the example that generated the feature (since each example can generate several features, as seen before);
- the offset mapping that will give us a map from token indices to character positions in the context.
That's why we will re-process the validation set with the following function, slightly different from prepare_train_features:
def prepare_validation_features(examples):
# Some of the questions have lots of whitespace on the left, which is not useful and will make the
# truncation of the context fail (the tokenized question will take a lots of space). So we remove that
# left whitespace
examples["question"] = [q.lstrip() for q in examples["question"]]
# Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
# in one example possible giving several features when a context is long, each of those features having a
# context that overlaps a bit the context of the previous feature.
tokenized_examples = tokenizer(
examples["question" if pad_on_right else "context"],
examples["context" if pad_on_right else "question"],
truncation="only_second" if pad_on_right else "only_first",
max_length=max_length,
stride=doc_stride,
return_overflowing_tokens=True,
return_offsets_mapping=True,
padding="max_length",
)
# Since one example might give us several features if it has a long context, we need a map from a feature to
# its corresponding example. This key gives us just that.
sample_mapping = tokenized_examples.pop("overflow_to_sample_mapping")
# We keep the example_id that gave us this feature and we will store the offset mappings.
tokenized_examples["example_id"] = []
for i in range(len(tokenized_examples["input_ids"])):
# Grab the sequence corresponding to that example (to know what is the context and what is the question).
sequence_ids = tokenized_examples.sequence_ids(i)
context_index = 1 if pad_on_right else 0
# One example can give several spans, this is the index of the example containing this span of text.
sample_index = sample_mapping[i]
tokenized_examples["example_id"].append(examples["id"][sample_index])
# Set to None the offset_mapping that are not part of the context so it's easy to determine if a token
# position is part of the context or not.
tokenized_examples["offset_mapping"][i] = [
(o if sequence_ids[k] == context_index else None)
for k, o in enumerate(tokenized_examples["offset_mapping"][i])
]
return tokenized_examples
And like before, we can apply that function to our validation set easily:
validation_features = datasets["validation"].map(
prepare_validation_features,
batched=True,
remove_columns=datasets["validation"].column_names
)
HBox(children=(FloatProgress(value=0.0, max=11.0), HTML(value='')))
Now we can grab the predictions for all features by using the Trainer.predict method:
raw_predictions = trainer.predict(validation_features)
The Trainer hides the columns that are not used by the model (here example_id and offset_mapping which we will need for our post-processing), so we set them back:
validation_features.set_format(type=validation_features.format["type"], columns=list(validation_features.features.keys()))
We can now refine the test we had before: since we set None in the offset mappings when it corresponds to a part of the question, it's easy to check if an answer is fully inside the context. We also eliminate very long answers from our considerations (with an hyper-parameter we can tune)
max_answer_length = 30
start_logits = output.start_logits[0].cpu().numpy()
end_logits = output.end_logits[0].cpu().numpy()
offset_mapping = validation_features[0]["offset_mapping"]
# The first feature comes from the first example. For the more general case, we will need to be match the example_id to
# an example index
context = datasets["validation"][0]["context"]
# Gather the indices the best start/end logits:
start_indexes = np.argsort(start_logits)[-1 : -n_best_size - 1 : -1].tolist()
end_indexes = np.argsort(end_logits)[-1 : -n_best_size - 1 : -1].tolist()
valid_answers = []
for start_index in start_indexes:
for end_index in end_indexes:
# Don't consider out-of-scope answers, either because the indices are out of bounds or correspond
# to part of the input_ids that are not in the context.
if (
start_index >= len(offset_mapping)
or end_index >= len(offset_mapping)
or offset_mapping[start_index] is None
or offset_mapping[end_index] is None
):
continue
# Don't consider answers with a length that is either < 0 or > max_answer_length.
if end_index < start_index or end_index - start_index + 1 > max_answer_length:
continue
if start_index <= end_index: # We need to refine that test to check the answer is inside the context
start_char = offset_mapping[start_index][0]
end_char = offset_mapping[end_index][1]
valid_answers.append(
{
"score": start_logits[start_index] + end_logits[end_index],
"text": context[start_char: end_char]
}
)
valid_answers = sorted(valid_answers, key=lambda x: x["score"], reverse=True)[:n_best_size]
valid_answers
[{'score': 16.706663, 'text': 'Denver Broncos'},
{'score': 14.635585,
'text': 'Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers'},
{'score': 13.234194, 'text': 'Carolina Panthers'},
{'score': 12.468662, 'text': 'Broncos'},
{'score': 11.709289, 'text': 'Denver'},
{'score': 10.397583,
'text': 'Broncos defeated the National Football Conference (NFC) champion Carolina Panthers'},
{'score': 10.104669,
'text': 'American Football Conference (AFC) champion Denver Broncos'},
{'score': 9.721636,
'text': 'The American Football Conference (AFC) champion Denver Broncos'},
{'score': 9.007437,
'text': 'Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers 24–10'},
{'score': 8.834958,
'text': 'Denver Broncos defeated the National Football Conference (NFC) champion Carolina'},
{'score': 8.38701,
'text': 'Denver Broncos defeated the National Football Conference (NFC)'},
{'score': 8.143825,
'text': 'Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers 24–10 to earn their third Super Bowl title.'},
{'score': 8.03359,
'text': 'American Football Conference (AFC) champion Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers'},
{'score': 7.832466,
'text': 'Denver Broncos defeated the National Football Conference (NFC'},
{'score': 7.650557,
'text': 'The American Football Conference (AFC) champion Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers'},
{'score': 7.6060467, 'text': 'Carolina Panthers 24–10'},
{'score': 7.5795317,
'text': 'Denver Broncos defeated the National Football Conference'},
{'score': 7.433568, 'text': 'Carolina'},
{'score': 6.742434,
'text': 'Carolina Panthers 24–10 to earn their third Super Bowl title.'},
{'score': 6.71136,
'text': 'Denver Broncos defeated the National Football Conference (NFC) champion Carolina Panthers 24'}]
We can compare to the actual ground-truth answer:
datasets["validation"][0]["answers"]
{'answer_start': [177, 177, 177],
'text': ['Denver Broncos', 'Denver Broncos', 'Denver Broncos']}
Our model picked the right as the most likely answer!
As we mentioned in the code above, this was easy on the first feature because we knew it comes from the first example. For the other features, we will need a map between examples and their corresponding features. Also, since one example can give several features, we will need to gather together all the answers in all the features generated by a given example, then pick the best one. The following code builds a map from example index to its corresponding features indices:
import collections
examples = datasets["validation"]
features = validation_features
example_id_to_index = {k: i for i, k in enumerate(examples["id"])}
features_per_example = collections.defaultdict(list)
for i, feature in enumerate(features):
features_per_example[example_id_to_index[feature["example_id"]]].append(i)
We're almost ready for our post-processing function. The last bit to deal with is the impossible answer (when squad_v2 = True). The code above only keeps answers that are inside the context, we need to also grab the score for the impossible answer (which has start and end indices corresponding to the index of the CLS token). When one example gives several features, we have to predict the impossible answer when all the features give a high score to the impossible answer (since one feature could predict the impossible answer just because the answer isn't in the part of the context it has access too), which is why the score of the impossible answer for one example is the minimum of the scores for the impossible answer in each feature generated by the example.
We then predict the impossible answer when that score is greater than the score of the best non-impossible answer. All combined together, this gives us this post-processing function:
from tqdm.auto import tqdm
def postprocess_qa_predictions(examples, features, raw_predictions, n_best_size = 20, max_answer_length = 30):
all_start_logits, all_end_logits = raw_predictions
# Build a map example to its corresponding features.
example_id_to_index = {k: i for i, k in enumerate(examples["id"])}
features_per_example = collections.defaultdict(list)
for i, feature in enumerate(features):
features_per_example[example_id_to_index[feature["example_id"]]].append(i)
# The dictionaries we have to fill.
predictions = collections.OrderedDict()
# Logging.
print(f"Post-processing {len(examples)} example predictions split into {len(features)} features.")
# Let's loop over all the examples!
for example_index, example in enumerate(tqdm(examples)):
# Those are the indices of the features associated to the current example.
feature_indices = features_per_example[example_index]
min_null_score = None # Only used if squad_v2 is True.
valid_answers = []
context = example["context"]
# Looping through all the features associated to the current example.
for feature_index in feature_indices:
# We grab the predictions of the model for this feature.
start_logits = all_start_logits[feature_index]
end_logits = all_end_logits[feature_index]
# This is what will allow us to map some the positions in our logits to span of texts in the original
# context.
offset_mapping = features[feature_index]["offset_mapping"]
# Update minimum null prediction.
cls_index = features[feature_index]["input_ids"].index(tokenizer.cls_token_id)
feature_null_score = start_logits[cls_index] + end_logits[cls_index]
if min_null_score is None or min_null_score < feature_null_score:
min_null_score = feature_null_score
# Go through all possibilities for the `n_best_size` greater start and end logits.
start_indexes = np.argsort(start_logits)[-1 : -n_best_size - 1 : -1].tolist()
end_indexes = np.argsort(end_logits)[-1 : -n_best_size - 1 : -1].tolist()
for start_index in start_indexes:
for end_index in end_indexes:
# Don't consider out-of-scope answers, either because the indices are out of bounds or correspond
# to part of the input_ids that are not in the context.
if (
start_index >= len(offset_mapping)
or end_index >= len(offset_mapping)
or offset_mapping[start_index] is None
or offset_mapping[end_index] is None
):
continue
# Don't consider answers with a length that is either < 0 or > max_answer_length.
if end_index < start_index or end_index - start_index + 1 > max_answer_length:
continue
start_char = offset_mapping[start_index][0]
end_char = offset_mapping[end_index][1]
valid_answers.append(
{
"score": start_logits[start_index] + end_logits[end_index],
"text": context[start_char: end_char]
}
)
if len(valid_answers) > 0:
best_answer = sorted(valid_answers, key=lambda x: x["score"], reverse=True)[0]
else:
# In the very rare edge case we have not a single non-null prediction, we create a fake prediction to avoid
# failure.
best_answer = {"text": "", "score": 0.0}
# Let's pick our final answer: the best one or the null answer (only for squad_v2)
if not squad_v2:
predictions[example["id"]] = best_answer["text"]
else:
answer = best_answer["text"] if best_answer["score"] > min_null_score else ""
predictions[example["id"]] = answer
return predictions
And we can apply our post-processing function to our raw predictions:
final_predictions = postprocess_qa_predictions(datasets["validation"], validation_features, raw_predictions.predictions)
Post-processing 10570 example predictions split into 10784 features.
HBox(children=(FloatProgress(value=0.0, max=10570.0), HTML(value='')))
Then we can load the metric from the datasets library.
metric = load_metric("squad_v2" if squad_v2 else "squad")
Then we can call compute on it. We just need to format predictions and labels a bit as it expects a list of dictionaries and not one big dictionary. In the case of squad_v2, we also have to set a no_answer_probability argument (which we set to 0.0 here as we have already set the answer to empty if we picked it).
if squad_v2:
formatted_predictions = [{"id": k, "prediction_text": v, "no_answer_probability": 0.0} for k, v in final_predictions.items()]
else:
formatted_predictions = [{"id": k, "prediction_text": v} for k, v in final_predictions.items()]
references = [{"id": ex["id"], "answers": ex["answers"]} for ex in datasets["validation"]]
metric.compute(predictions=formatted_predictions, references=references)
{'exact_match': 76.74550614947965, 'f1': 85.13412652023338}
You can now upload the result of the training to the Hub, just execute this instruction:
trainer.push_to_hub()
You can now share this model with all your friends, family, favorite pets: they can all load it with the identifier "your-username/the-name-you-picked" so for instance:
from transformers import AutoModelForQuestionAnswering
model = AutoModelForQuestionAnswering.from_pretrained("sgugger/my-awesome-model")