Slot filling model

[rlucatoor]

What is Slot Filling

Slot Filling is the task to extract specific pieces of information (slots) from a sentence. For instance, given the sentence

I want to fly from Boston at 8:38 am and arrive in Denver at 11:10 in the morning

and the slots departure_city, arrival_city, departure_time and arrival_time, a Slot Filling model would return the following output

{
  "departure_city": "Boston",
  "departure_time": "8:38 am",
  "arrival_city": "Denver",
  "arrival_time": "11:10 am"
}

What needs to be done

Implement a Slot Filling model that exposes the following public methods:

.train() method:

.train(slots: list[str], output_path: Optional[str] = None, num_samples: int = config.DEFAULT_SYNTHEX_DATAPOINT_NUM, num_epochs: int = 3)`

where the slots argument accepts a list of strings with the following constraints:

1. Maximum 20 characters
2. No spaces

The method should then train the model to identify the provided slots in any kind of sentence. The BaseModel._generate_synthetic_data() method should be used to generate synthetic training data.

.__call__(text: Union[str, list[str])

• If a str is provided, the model should

  • Extract the slots it was trained for from the string
  • Return a dictionary, where the keys are the slot names and the values are the extracted pieces of text. If a given slot is not found in the string, its value in the dictionary should be None.

• If a list[str] is provided, the model should return a list of dictionaries, one per sentence.

.load(model_path: str)

Loads a pre-trained Slot Filling model from the specified path.