LLM Lipogram Challenge Portal

Exercise Instructions

Exercise 1: La disparition (No 'e' or 'E')

Generate text without ever using the letter 'e' or 'E'. For this, you must use model() directly: model(input_ids) yields logits. You need to manually adjust the logits to forbid tokens containing 'e' or 'E'. REQUIREMENT: Do NOT use model.generate().

Exercise 2: The Toulouse Sequence

Generate text without ever using the word 'Toulouse'. For this, you must use model() directly: model(input_ids) yields logits. You need to manually adjust the logits. It is more difficult here because 'Toulouse' is a multi-token word. REQUIREMENT: Do NOT use model.generate().

Download the challenge.py template below to get started: