Projects

• Piaget

  Piaget is a family of Qwen3 models, from 0.6B to 8B parameters, adapted for psychological and philosophical reasoning.

  I built PsychologicalReasoning-15k by selecting psychology- and philosophy-related reasoning traces from large open corpora using embeddings, clustering, and domain classification.

• Beck

  Beck is a family of Piaget-based models, from 0.6B to 8B parameters, further adapted through preference optimization on psychotherapeutic data.

• Oneirogen

  Oneirogen is a language model fine-tuned on more than 27,000 DreamBank narratives to generate dream accounts. I developed it alongside my work on extracting characters and emotions from dreams.

• Dream-T5

  Dream-T5 is a sequence-to-sequence model for extracting characters and their emotions from dream narratives. It implements the method introduced in my LREC-COLING 2024 paper and was used to produce the accompanying DreamBank annotations.

• French emotion models

  CamemBERT models for identifying components of emotional experience and predicting discrete emotions in French autobiographical narratives. Released with the corpus from our EACL 2023 workshop paper.

• Preference optimization with LoRA and ORPO

  Code for preference optimization with ORPO and LoRA, without full-model fine-tuning.