ALMAnaCH's fully-open bilingual FR-EN-code generative LLM suite (1.5B/8B/24B, 2–4T tokens), released with the full data-curation pipeline and hundreds of checkpoints — the Inria-led open-LLM answer, extending the CamemBERT lineage from encoders into generative models. Distinctive research angle: deliberately "peppered" harmless data-poisoning as a built-in safety testbed.

Model Details

Parameters 24B
Training tokens 4T

Variants

Name Parameters Notes
Gaperon-8B 8B
Gaperon-1.5B 1.5B
Gaperon-24B 24B

Paper

open-weightmultilingualpretraining

Related