Model Card for Luciole-8B-KD-Math-Thinking-1.1

• Model Description
  • Bias, Risks, and Limitations
  • Recommendations
• Training Details
  • Training Data
  • Instruction template
  • Training Procedure
• Evaluation
• Testing the model
  • Test with ollama
  • Test with vLLM
• Citation
• Acknowledgements
• Contact

NOTE: Although this model has undergone safety alignment during the DPO thinking phase, it could still generate unsafe output for certain inputs. The model also has not been trained on identity prompts, so it may not correctly answer the questions about its name/origin. The Luciole models are under active development and we plan to address these issues in future versions.

Model Description

Luciole-8B-KD-Math-Thinking-1.1 is a fine-tuned and aligned version of Luciole-8B-KD-Math-Base, an open-source, multilingual causal language model created by OpenLLM-France. Luciole-8B-KD-Math-Thinking-1.1 was developed by LINAGORA and the OpenLLM-France consortium as a part of the OpenLLM France project, funded by BPI France through the France 2030 program.

Training of Luciole-8B-KD-Math-Thinking-1.1 was conducted on GENCI supercomputer Jean Zay, managed by IDRIS. It was trained in two phases: (i) a supervised fine-tuning (SFT) phase on instruction data with thinking traces, and (ii) a final preference alignment phase using Direct Preference Optimization (DPO). The training data covers topics in math, science, coding, general chat, RAG and translation.

Note that Luciole-8B-KD-Math-Thinking-1.1 is only the first iteration of post-trained models based on Luciole-8B-KD-Math-Base. Development of the Luciole models is an active, ongoing project and further iterations, including models trained on higher proportions of French data, are planned. In the spirit of open-source, we share the model weights, training recipes, and the training data to facilitate research and to provide open source building blocks for other training projects.

If you are interested in contributing to the Luciole project, contact us at contact@openllm-france.fr.

• License: Apache 2.0
• Training repository: Luciole-Training
• Training data: Luciole-PostTraining-Dataset-1.1
• Technical report: coming soon.

Bias, Risks, and Limitations

Luciole-8B-KD-Math-Thinking-1.1 is the result of a first phase of fine-tuning and alignment to human preferences. Efforts to improve the model are active and ongoing and the model should be thoroughly tested for target use cases before being incorporated in industrial pipelines. Training has mostly focused on general instruction following, and additional training would be necessary for tasks specifically focused on code generation and mathematical problem solving. While the model has seen some tool calling data, it has a tendency to over-refuse tool calling requests. These capacities are experimental and will be improved in future versions.

Luciole-8B-KD-Math-Thinking-1.1 was post-trained almost entirely on English data. Future post-training phases will focus specifically on increasing the proportion of French data and testing the impact that this has on model performance.

Due to its size, Luciole-8B-KD-Math-Thinking-1.1 is limited in the information that it can memorize. It can also hallucinate; that is, produce false answers that result from its training on massive amounts of diverse text. Its ability to produce correct answers could be improved by implementing the model in a retrieval augmented generation pipeline.

Finally, Luciole-8B-KD-Math-Thinking-1.1 was trained on sequences of 16,384 tokens.

Recommendations

• Further train Luciole-8B-KD-Math-Thinking-1.1 for specific use cases.
• Integrate the model in a RAG pipeline to augment its knowledge base.
• Extend training with longer sequences to increase context length.

Training details

Training data

The post-training data and dataset description is available here.

Instruction template

Luciole-8B-KD-Math-Thinking-1.1 was trained on the chat template inspired from Qwen3-1.7B. In our chat template, we use the default system prompt "You are a helpful AI assistant named Luciole, trained by LINAGORA and OpenLLM France.".

As an example, when applying the chat template to the user query below:

from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1")
query = [
  {
    "content": "Who was Molière?",
    "role": "user"
  }
]
print(tokenizer.apply_chat_template(query, tokenize=False, add_generation_prompt=True))

we get the following formatted prompt:

<|im_start|>system
You are a helpful AI assistant named Luciole, trained by LINAGORA and OpenLLM France.<|im_end|>
<|im_start|>user
Who was Molière?<|im_end|>
<|im_start|>assistant

Given this input, the model first generates a thinking trace, enclosed in thinking tags, before giving its final response:

<think>
Okay, the user is asking who Molière was, and they want a concise answer. Let me start by recalling what I know about Molière …
</think>

Jean-Baptiste Poquelin, known as Molière, was a 17th-century French playwright, actor, and poet. Born in Paris in 1622, … <|im_end|>

Thinking can be turned off by using tokenizer.apply_chat_template(…, enable_thinking=False). This will append empty thinking tags (<think>\n\n</think>\n\n) in the prompt, immediately after "assistant\n". However, for testing a non-thinking model, it is better to conduct experiments on Luciole-8B-Instruct-1.1.

By default, the <think>…</think> reasoning of all previous assistant turns is kept when formatting a multi-turn conversation. Setting tokenizer.apply_chat_template(…, preserve_thinking=False) drops it from every assistant turn preceding the last user message (only the most recent turn keeps its reasoning), which shortens the prompt and reduces the KV-cache footprint. With vLLM, set it per request via extra_body={"chat_template_kwargs": {"preserve_thinking": false}}. Ollama does not use this parameter, as it formats prompts with its own built-in renderer.

Training procedure

Luciole-8B-KD-Math-Thinking-1.1 was trained in two phases. The model was first finetuned on thinking data, followed by DPO on preference data with thinking traces.

	SFT with thinking	DPO with thinking
Num training samples	2.6M	284K
Num epochs	4	1
Max learning rate	2.0e-5	3.0e-8
Min learning rate	2.0e-6	0
LR scheduler	cosine	linear
Max sequence length	16384	16384
Batch size	2048	128
DPO β	n/a	5

Links to the interim models: SFT with thinking.

Evaluation

The two figures compare the performance of Luciole-8B-KD-Math-Thinking-1.1 with other thinking models on English and French benchmarks. All the models were evaluated with temperature = 0 and a thinking budget of 5000 tokens. For the Olmo models, we ran evaluation on Think-DPO variant as Olmo-Think had undergone an additional stage of RLVR training.

Apart from Olmo-3-32B-Think-DPO and Qwen-3-8B, Luciole-8B-KD-Math-Thinking-1.1 has the best overall performance on the English benchmarks. In particular, Luciole-8B-KD-Math-Thinking-1.1 clearly outperforms Olmo-3-7B-Think-DPO, and Ministral-3-8B-Reasoning-2512 on instruction following benchmarks such as IFEval, IFBench test, IFBench Multiturn.

Luciole-8B-KD-Math-Thinking-1.1 displays strong performance on the French benchmarks tested. The overall performance of the model is at par with Olmo-3-32B-Think-DPO. Luciole-8B-KD-Math-Thinking-1.1 performs particularly well on benchmarks like EIFFEL (a French cultural benchmark), INCLUDE fr gen, BBH fr, and IFEval-fr. Note that Luciole-8B-KD-Math-Base is distilled from a corrected version of Luciole-23B-Base, which was trained on roughly 30% French data. This likely explains the strong performance of Luciole-8B-KD-Math-Instruct-1.1 on French benchmarks.

Testing the model

Test with ollama

⚠️ While we are publishing the GGUF model, it is currently impossible to use the model with ollama due to a bug in llama.cpp (https://github.com/ggml-org/llama.cpp/issues/30518). Once this issue is fixed, the model will be usable.

Note: for best performance, please use the GGUF files released officially by OpenLLM.

• Download and install Ollama
• Download the GGUF model
• Copy the Modelfile, adpating if necessary the path to the GGUF file (line starting with FROM).
• Run in a shell:
  • ollama create -f Modelfile Luciole
  • ollama run Luciole
• Once ">>>" appears, type your prompt(s) and press Enter.
• Optionally, restart a conversation by typing "/clear"
• End the session by typing "/bye".

Useful for debug:

• How to print input requests and output responses in Ollama server?
• Documentation on Modelfile
  • Examples: Ollama model library
    • Llama 3 example: https://ollama.com/library/llama3.1
• Add GUI : https://docs.openwebui.com/

Test with vLLM

1. Run vLLM Docker Container

Use the following command to deploy the model, replacing INSERT_YOUR_HF_TOKEN with your Hugging Face Hub token.

docker run --runtime nvidia --gpus=all \
    --env "HUGGING_FACE_HUB_TOKEN=INSERT_YOUR_HF_TOKEN" \
    -p 8000:8000 \
    --ipc=host \
    vllm/vllm-openai:latest \
    --model OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1

2. Test using OpenAI Client in Python

To test the deployed model, use the OpenAI Python client as follows:

from openai import OpenAI

# Initialize the client
client = OpenAI(base_url='http://localhost:8000/v1', api_key='empty')

# Define the input content
content = "Hello Luciole"

# Generate a response
chat_response = client.chat.completions.create(
    model="OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1",
    messages=[
        {"role": "user", "content": content}
    ],
)
print(chat_response.choices[0].message.content)

Citation

✍ Paper coming soon!

Acknowledgements

We gratefully acknowledge BPI France for funding the OpenLLM France project under the call "Communs numériques pour l’intelligence artificielle générative" ("Digital commons for generative artificial intelligence") as a part of the France 2030 program.

Training of Luciole-8B-KD-Math-Thinking-1.1 was made possible by computing AI and storage resources by GENCI at IDRIS thanks to the project A0181016189 on the supercomputer Jean Zay’s H100 partition. We gratefully acknowledge their support.

Luciole-8B-KD-Math-Thinking-1.1 was created by members of LINAGORA and the OpenLLM-France community, including in alphabetical order:

Akshay Chaturvedi (LINAGORA)
Liam Duignan (CEA List)
Olivier Ferret (CEA List)
Olivier Gouvert (LINAGORA)
Émile Hazard (Opsci)
Julie Hunter (LINAGORA)
Jean-Pierre Lorré (LINAGORA)
Jérôme Louradour (LINAGORA)
Michel-Marie Maudet (LINAGORA)
Kate Thompson (LINAGORA)
Matteo Van Ypersele de Strihou (LINAGORA)

Particular thanks to the following OpenLLM France members for their valuable input: Niyar Barman (LINAGORA/IRIT) for training OpenLLM-France/Luciole-8B-KD-Math-Base during his internship at LINAGORA, Clément Bénesse (Opsci), Christophe Cerisara (LORIA) and Gabriel Lauzzana (LORIA).

We thank the support teams from IDRIS and NVIDIA for technical guidance throughout the project.

We would also like to thank members of the Gaperon and Salamandra projects for sharing their insights with us. We also acknowledge the numerous open source actors whose resources have guided us throughout the training process, with particular thanks to Nvidia, HuggingFace and Allen AI.

Finally, we thank the entire OpenLLM-France community, whose members have helped in diverse ways.

Contact

contact@openllm-france.fr