SOTA NLP Workshop @ YouScan

For the last year there was a big progress in NLP, and now it is easier to build a good prototype for your product. We will show how to solve three tasks really quickly with a small effort.

We will use transformers library under the hood.

Set virtual environment

1. Create virtual env with python3.6 -m venv venv
2. Activate environment with source venv/bin/activate
3. Install all packages with pip install -r requirements.txt

Language generation

Simple language generation with GPT-2 model.

We provide command line interface with the next params:

• main parameters:

  • input_text – starting point of our generation
  • n_sentences – number of sentences to generate

• technical:

  • temperature – technical parameter to add some randomness
  • repetition_penalty – penalize model for repetition
  • top_k – choose prediction from k most probable tokens
  • top_p – choose prediction from the top tokens with cumulative probability >= top_p

Run python language_generation.py --input_text="Write some text"

Choose the best answer

The task is to find the best answer from the list to the asked question.

We provide command line interface with the next params:

• question – the question, we must find the answer to
• file_path – file path to file with answers list

Questions:

• How are you?
• How old are you?
• What is your job?
• Did you enjoy last trip?
• Was it an improvisation?

Run python choose_best_answer.py --question="How are you?"

Find the answer in the paragraph

The task is to find the answer to the question based on the provided document.

We provide command line interface with the next params:

• question – the question, we must find the answer to
• file_path – file path to a paragraph

Questions:

• When was he born?
• What is his profession?
• How many goals did he score in Champions League?
• Where does Ronaldo play now?
• What trophies did Ronaldo win?
• How many goals did he score?

Run python find_answer_in_paragraph.py --question="What is his profession?"