---

Luth-1.7B-Instruct

Luth-1.7B-Instruct is a French fine-tuned version of Qwen3-1.7B, trained on the Luth-SFT dataset. The model has drastically improved its French capabilities in instruction following, math, and general knowledge. Additionally, its English capabilities have remained stable and have even increased in some areas.

Our Evaluation, training and data scripts are available on GitHub, along with the Blog we wrote.

Model Details

Luth was trained using full fine-tuning on the Luth-SFT dataset with Axolotl. The resulting model was then merged with the base Qwen3-1.7B model. This process successfully retained the model's English capabilities while improving its performance on most selected benchmarks in both French and English.

Benchmark Results

We used LightEval for evaluation, with custom tasks for the French benchmarks. The models were evaluated with a temperature=0.

Evaluation Visualizations

French Evaluation:

English Evaluation:

French Benchmark Scores

Benchmark	Qwen3-1.7B	SmolLM2-1.7B-Instruct	Qwen2.5-1.5B-Instruct	Luth-1.7B-Instruct
ifeval-fr	54.53	31.24	32.90	57.67
gpqa-diamond-fr	26.90	21.83	28.93	38.58
mmlu-fr	28.46	33.73	46.25	49.66
math-500-fr	60.80	11.20	32.20	64.00
arc-chall-fr	33.28	28.57	32.68	35.16
hellaswag-fr	24.86	49.58	34.34	31.93

English Benchmark Scores

Benchmark	Qwen3-1.7B	SmolLM2-1.7B-Instruct	Qwen2.5-1.5B-Instruct	Luth-1.7B-Instruct
ifeval-en	68.39	48.24	39.93	65.80
gpqa-diamond-en	31.82	24.75	30.30	31.82
mmlu-en	52.74	50.27	59.81	60.19
math-500-en	69.20	22.40	56.00	70.00
arc-chall-en	36.09	42.32	41.04	42.24
hellaswag-en	46.96	66.94	64.48	58.55

Code Example

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("kurakurai/Luth-1.7B-Instruct")
model = AutoModelForCausalLM.from_pretrained("kurakurai/Luth-1.7B-Instruct")
messages = [
    {"role": "user", "content": "Quelle est la capitale de la France?"},
]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=100)
print(
    tokenizer.decode(
        outputs[0][inputs["input_ids"].shape[-1] :], skip_special_tokens=True
    )
)

Citation

@misc{luth2025kurakurai,
  title   = {Luth-1.7B-Instruct},
  author  = {Kurakura AI Team},
  year    = {2025},
  howpublished = {\url{https://huggingface.co/kurakurai/Luth-0.6B}},
  note    = {Qwen3-1.7B fine-tuned on French datasets}
}