Artificial intelligence models for reading documents and answering questions
File: document_qa_demo.ipynb
This file is mainly used to learn some functions in pipline, such as calling some pre-trained model, model-to-input transformations, e.g. 'tensors' and 'logits'. They are the main recognition types in Transformers.
raw_inputs = [
"I enjory my intership time in BOPRC.",
"I like it so much!",
]
inputs = tokenizer(raw_inputs, padding=True, truncation=True, return_tensors="tf")
print(inputs)Here, any two sentences can be written in raw_inputs, with some forward or reverse expression. After that the inputs returned to me processed by the 'AutoTokenizer' model with distilbert-base-uncased-finetuned-sst-2-english as a checkpoint look like this:
{'input_ids': <tf.Tensor: shape=(2, 14), dtype=int32, numpy=
array([[ 101, 1045, 4372, 5558, 2854, 2026, 6970, 9650, 2051,
1999, 29432, 11890, 1012, 102],
[ 101, 1045, 2066, 2009, 2061, 2172, 999, 102, 0,
0, 0, 0, 0, 0]], dtype=int32)>,
'attention_mask': <tf.Tensor: shape=(2, 14), dtype=int32, numpy=
array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0]], dtype=int32)>}