Extend graphs

If you would like to extend existing dialog graph, LLMGraphExtender is a class to use. This class aims to invoke LLMs for increasing node and edge number in a dialog graph.

First, load your data and parse it with RawDGParser.

from datasets import load_dataset
from dialog2graph.pipelines.helpers.parse_data import RawDGParser, PipelineRawDataType

dataset = load_dataset("DeepPavlov/d2g_generated", token=True)
data_example = dataset['train'][0]
data_example = PipelineRawDataType(dialogs=data_example['dialogs'], supported_graph=data_example['graph'])

parser = RawDGParser()
parsed_data = parser.invoke(data_example)

Then, you can pass the data to LLMGraphExtender and get extended graph. Set enable_evals to True to get metric report (it may slow down response time).

from dialog2graph.pipelines.model_storage import ModelStorage
from dialog2graph.pipelines.d2g_extender.three_stages_extender import LLMGraphExtender

ms = ModelStorage()
graph_extender = LLMGraphExtender(model_storage=ms)