BlenderBotSmall-News: Small version of a state-of-the-art open source chatbot, trained on custom summaries

Details of BlenderBotSmall

The BlenderBotSmall model was presented in A state-of-the-art open source chatbot by Facebook AI and here are it's details:

Details of the downstream task (Summarization) - Dataset 📚

A custom dataset was used, which was hand prepared by SmokeTrees Digital AI engineers. This data contains long texts and summaries.

Model training

The training script is present here.

Pipelining the Model

model = transformers.BlenderbotSmallForConditionalGeneration.from_pretrained('lordtt13/blenderbot_small-news')

tokenizer = transformers.BlenderbotSmallTokenizer.from_pretrained("lordtt13/blenderbot_small-news")

nlp_fill = transformers.pipeline('summarization', model = model, tokenizer = tokenizer)
nlp_fill('The CBI on Saturday booked four former officials of Syndicate Bank and six others for cheating, forgery, criminal conspiracy and causing ₹209 crore loss to the state-run bank. The accused had availed home loans and credit from Syndicate Bank on the basis of forged and fabricated documents. These funds were fraudulently transferred to the companies owned by the accused persons.', min_length=5, max_length=40)

# Output:
# [{'summary_text': 'marize: the cbi booked four former officials of syndicate bank and six others for cheating , forgery , criminal conspiracy and causing 209 crore loss to the staterun bank'}]

Created by Tanmay Thakur | LinkedIn