metadata

widget:
  - text: mammal is to whale what
    example_title: Analogy Example 1 (semantic relation)
  - text: 'wedding is to marriage what '
    example_title: Analogy Example 2 (semantic relation, metaphor)
  - text: London is to U.K. what
    example_title: Analogy Example 3 (entity)
  - text: actual is to actually what
    example_title: Analogy Example 4 (morphological)

relbert/opt-350m-analogy

This is facebook/opt-350m fine-tuned on relbert/semeval2012_relational_similarity for analogy generation, which is to generate a word pair (eg. bird is to crow) given a query (eg. mammal is to whale) so that the query and the generated word pair form an analogy statement.

Usage

from transformers import pipeline

pipe = pipeline('text2text-generation', model="relbert/opt-350m-analogy")
output = pipe("mammal is to whale what")
print(output)
>>> [{'generated_text': 'bird is to crow'}]