Alpha
Contact
Implement · Luciole 1.1 Instruct

Can you run it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input D ยท 51.2/100

This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Install & run

The 23B and 1B load with transformers through the native Nemotron class, so no repository code is needed. Pin a revision and check the shard hashes against the Hugging Face LFS values (ev-card-instruct, ev-hf-platform).

from transformers import AutoTokenizer, AutoModelForCausalLM
name = "OpenLLM-France/Luciole-23B-Instruct-1.1"
tok = AutoTokenizer.from_pretrained(name)   # pin revision=<commit sha>
model = AutoModelForCausalLM.from_pretrained(name, torch_dtype="bfloat16", device_map="auto")

For Ollama, download an official GGUF file and the Modelfile from the GGUF repo, edit the FROM path, then run ollama create -f Modelfile Luciole. The Modelfile sets a 32,000-token context and temperature 0.2.

The 8B repo ships configuration_nemotron_h.py and modeling_nemotron_h.py. Read them before executing and run them in a sandbox (ev-hf-platform).

Hardware & VRAM requirements

The publisher gives file sizes and no requirements table (ev-hf-platform, ev-card-instruct).

Artefact23B8B1B
bf16 safetensors46.4 GBnot measured2.6 GB
GGUF BF1646.4 GB16.2 GB3.2 GB
GGUF Q8_024.7 GB8.6 GB1.7 GB
GGUF Q5_K_M16.7 GB5.8 GB1.2 GB
GGUF Q4_K_M14.5 GB5.0 GB1.0 GB
FP8 (compressed-tensors)24.8 GBlisted in the orglisted in the org

Add memory for the KV cache. A 131k context needs far more memory than the weights alone, and the publisher gives no figure. FP8 needs Hopper, Ada Lovelace or Blackwell GPUs.

Serving stacks

StackStatus
vLLMdocumented with --enable-auto-tool-choice --tool-call-parser hermes; the Thinking card needs a recent vLLM (0.24 or later) and --reasoning-parser qwen3
Ollamaofficial GGUF and Modelfile
llama.cppofficial GGUF, parsed by Hugging Face as architecture nemotron with a 131,072 context
transformersnative Nemotron class for 23B and 1B; repository code for the 8B hybrid
Hosted providersnone; the Hugging Face inference provider mapping is empty

Safe-deployment controls & Deployment Ceiling

Applying the safe-deployment playbook:

  • Supply chain: pin the revision, verify the LFS hashes of every shard, and add the Apache-2.0 text and the LINAGORA SAS notice to anything you redistribute.
  • Model layer: no safety evaluation, acceptable-use policy or guard model ships. Run your own misuse tests in each target language, since post-training was almost entirely English.
  • Tool calling: the card marks it as work in progress. Test it on your tools before use.
  • Input and output layers: add your own guardrails and logging.
  • Evaluation: the only published results are the publisher's instruction-following and maths benchmarks.

Deployment Ceiling: T1 (internal assistant, human reads every output), conditional. No safety evidence supports customer-facing T2 use. The ceiling moves up when a safety evaluation is published and your own tests pass.

Available quantizations

OpenLLM-France publishes GGUF (BF16, Q8_0, Q5_K_M, Q4_K_M) and FP8 builds of the 1B, 8B and 23B Instruct models, all Apache-2.0 under the same copyright notice (ev-siblings). The card says to prefer the official GGUF files for Ollama. No quality-loss figures are published, so test Q4_K_M on your task.

Fine-tuning & adaptation

The card recommends further training for specific use cases and longer-sequence training to extend context (ev-card-instruct). The training repository holds NeMo and NeMo-RL recipes and per-model cards for the Instruct and Thinking runs (ev-code-repo). The SFT and DPO data is public, with per-source tables (ev-posttrain-data).

The published recipe is SFT on 2.6M thinking samples and 2.1M non-thinking samples for four epochs at a 2e-5 peak learning rate, then DPO on 284K pairs with beta 5. The recipes target the Jean Zay cluster and Slurm, so expect to adapt them. The finetune folder has no README on the main branch.

API / OpenAI-compatible integration

vLLM serves an OpenAI-compatible endpoint. The card shows a chat request with the OpenAI Python client and a tool-calling request with a tools list. Calls return in message.tool_calls (ev-card-instruct). Ollama exposes its own API.

How this scores

The ownership factors this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Weak

Operational readiness is 3, and safety is 1 because no safety evaluation, guard model or acceptable-use policy is published. The Thinking card states that safety alignment can still leave unsafe output (ev-card-instruct, ev-card-thinking).

How this scores (AOI sub-dimensions)
Operational3/5how practical it is to run, serve and maintain in productionvLLM, Ollama and llama.cpp routes are documented, with official GGUF and FP8 builds.
Safety1/5whether misuse risks are evaluated and guardrails are providedThe Instruct card has no safety section, no acceptable-use policy and no guard model.
4

Doesn't extract your dataDoes running it keep your knowledge and data yours?

Strong

The weights run on your own hardware under Apache-2.0 with no clawback clause. The 23B repo ships only safetensors, tokenizer and configuration files, with no loader code (ev-card-instruct, ev-hf-platform).

How this scores
Not a scored AOI dimension. For a self-hosted model, data-control is a structural property of running the weights yourself, strong by default unless the model phones home or the licence claws back rights. For a hosted API this factor is the retention + train-on-inputs + residency read, scored from the binding terms.
What this means for adoptionYou own the use of the weights. Apache-2.0 is declared, the repos are ungated and your data stays on your hardware. Training code and both datasets are public, so most of what the model is can be inspected. You cannot rely on its safety, because no safety evaluation or guard model exists. The level rises when a safety evaluation, a LICENSE file in the Instruct repos and a technical report are published.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-11-19
The 23B Instruct card declares Apache-2.0 and names LINAGORA SAS as copyright holder.
Model cardread2026-11-19
The Thinking-1.1 card opens with a note that the model 'has undergone safety alignment during the DPO thinking phase' and can still produce unsafe output.
Model cardread2026-11-19
The Luciole-23B-Base card declares Apache-2.0 and lists French, English, Italian, German, Spanish, Portuguese, Dutch and Arabic with token shares.
Licenceread2026-11-19
The Base repos contain a LICENSE.md holding the full Apache License 2.0 text with the template front matter and an unfilled 'Copyright [yyyy] [name of copyright owner]' line.
Licenceread2026-11-19
The NOTICE file of the Luciole-Training repository says the Luciole models, including Instruct-1.1 in all three sizes, are the property of LINAGORA SAS and released under Apache-2.0.
Licenceread2026-11-19
The LICENSE file of the Luciole-Training repository is the standard GNU General Public License version 3 text.
Documentationread2026-11-19
The Luciole-Training repository has 532 commits and holds data processing, tokenizer training, NeMo pretraining, conversion, NeMo and NeMo-RL fine-tuning folders and a lighteval-based evaluation folder.
Documentationread2026-11-19
The OpenLLM-France organisation is not verified on Hugging Face, has 20 members and lists 38 model repos.
Documentationread2026-11-19
Luciole-PostTraining-Dataset-1.1 is public and ungated under CC-BY-SA-4.0.
Documentationread2026-11-19
Luciole-Training-Dataset is public and ungated under CC-BY-SA-4.0, with per-source licences listed.
Model cardread2026-11-19
The 8B and 1B Instruct cards match the 23B card.
Technical_reportunverified2026-11-19
No Luciole technical report exists.
Model cardread2026-11-19
The benchmark figure on the 23B Instruct repo, made by the publisher, compares Luciole with Gaperon-1125-24B-SFT, EuroLLM-22B-Instruct-2512 and two Mistral-Small-24B models.
Vendor announcementread2026-11-19
The consortium site calls the project 'totalement open source', says it is funded by Bpifrance for two years with nine partners, and says it republishes training datasets.
Documentationread2026-11-19
The Nemotron-Instruction-Following-Chat-v1 card says its data came from GPT-OSS-120B and Qwen3-235B and is CC-BY-4.0.
Documentationread2026-11-19
RedTeachingBench is a French false-premise benchmark with 1,000 prompts for teachers using LLMs, under CC-BY-SA-4.0.