Lucie 7B

weights

open(Apache-2.0)

data

open(Lucie-Training-Dataset on HF, redistributed in training form)

code

open(Lucie-Training Megatron-DeepSpeed fork)

checkpoints

open(intermediate HF revisions)

license

Apache-2.0(OSI)

Full open-source tier alongside OLMo / Apertus: open weights, open training data, open training code, intermediate checkpoints. Authors position Lucie as OSI-definition-compliant for open-source AI. Training-dataset product deferred — artifacts cited here for openness only. Caveat found on the 2026-07-30 re-read: Lucie-Training-Dataset carries `cc-by-nc-sa-4.0`, so the corpus is published and ungated but NOT commercially reusable. The `data` dimension asks whether the corpus is released, not under what terms, so the score is unchanged — but the authors' OSI-compliance claim does not survive a strict reading, and Lucie is a weaker full-openness exemplar than OLMo, whose Dolma is `odc-by`.

• https://huggingface.co/api/datasets/OpenLLM-France/Lucie-Training-Dataset recorded 2026-07-30

  `gated: false`, `private: false`, 2,823 downloads, 38 likes. Card license is `cc-by-nc-sa-4.0` - the corpus is published and ungated, but non-commercial. See the note.

• https://huggingface.co/OpenLLM-France/Lucie-7B recorded 2026-07-30

  License apache-2.0, safetensors weights, links to Lucie-Training-Dataset, and intermediate checkpoints as repo branches (`step0005000`, `step0010000`, ...); `isGated: false`.

• https://huggingface.co/datasets/OpenLLM-France/Lucie-Training-Dataset recorded 2026-07-30

  Dataset page for the multilingual pretraining corpus: `isGated: false`, `isPrivate: false`, Parquet format present. Cited previously WITHOUT the `/datasets/` path segment, which the Hub resolves as a model repo and answers 401; the URL was corrected on re-read rather than the dataset having become gated.

• https://github.com/OpenLLM-France/Lucie-Training recorded 2026-07-30

  Repository described as "Code for continual pretraining of LUCIE" - the Megatron-DeepSpeed fork used for the run.

• https://arxiv.org/abs/2503.12294 recorded 2026-07-30

  Paper claiming OSI-compliant open resources for multilingual generation. The claim is the authors', which is why it establishes no dimension by itself.

Niche sovereign/French open-LLM audience. HF Lucie-7B ~780 downloads/month and 31 likes (Jul 2026); collection upvotes modest. Comparable adoption tier to Apertus / Falcon-class fully-open European releases rather than Llama/Qwen scale.

• https://huggingface.co/OpenLLM-France/Lucie-7B recorded 2026-07-19

  ~780 downloads/month, 31 likes (as of 2026-07-19)

• https://huggingface.co/collections/OpenLLM-France/lucie-llm recorded 2026-07-19

  official Lucie LLM collection (base, instruct, dataset, paper)

7B multilingual base prioritizing French cultural representation and data-rights constraints — competitive for its size/language mix on French/English evals in the paper, but well below the 2026 open-weight frontier (trillion-param MoEs). Capability 2 reflects small-scale fully-open sovereign model, not frontier performance. Value is openness + FR/EU language coverage.

• https://arxiv.org/abs/2503.12294 recorded 2026-07-19

  Lucie-7B / Instruct eval narrative vs contemporary open baselines

• https://huggingface.co/OpenLLM-France/Lucie-7B recorded 2026-07-19

  training/eval learning curves and multilingual benchmark plots on the model card