Instructions to use legmlai/legml-v1.0-8b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use legmlai/legml-v1.0-8b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="legmlai/legml-v1.0-8b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("legmlai/legml-v1.0-8b")
model = AutoModelForCausalLM.from_pretrained("legmlai/legml-v1.0-8b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use legmlai/legml-v1.0-8b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "legmlai/legml-v1.0-8b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "legmlai/legml-v1.0-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/legmlai/legml-v1.0-8b

SGLang

How to use legmlai/legml-v1.0-8b with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "legmlai/legml-v1.0-8b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "legmlai/legml-v1.0-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "legmlai/legml-v1.0-8b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "legmlai/legml-v1.0-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use legmlai/legml-v1.0-8b with Docker Model Runner:
```
docker model run hf.co/legmlai/legml-v1.0-8b
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

legml-v1.0-base — L’Excellence Française du Pre-Training Continu

L'IA pure qui forme l'IA : un corpus 100 % francophone sélectionné et contrôlé
Curated by legml.ai – Leader in AI Data Curation & Quality Assurance

1 • Présentation

legmlai/legml-v1.0-base est un grand modèle de langage (LLM) francophone basé sur Qwen-3 · 8 B.
Il bénéficie d’un pré-entraînement continu (continued pre-training) visant à renforcer la cohérence et la richesse contextuelle de ses réponses en français.

Projet créé et maintenu par Mohamad Alhajar.

🙏 Merci à Nebius pour le sponsoring GPU : 24 × H100 80 Go qui ont permis cette release.

2 • Versions disponibles

Modèle	Usage principal	Checkpoint	Héritage
legml-v1.0-base	Foundation — pré-entraînement continu	`legmlai/legml-v1.0-base`	Qwen-3 8 B
legml-v1.0-instruct	Dialogue & instruction-tuning	`legmlai/legml-v1.0-instruct`	Fine-tuning sur v1.0-base

3 • Fiche technique de v1.0-base

Paramètre	Valeur
Architecture	Transformer décodant (Qwen-3 8B), Rotary PE, FlashAttention-2
Taille du modèle	~ 16 Go (fp16) / 8 Go (bf16)
Vocabulaire	151 k tokens (compat. Qwen)
Corpus	≈ 1.2 T tokens, dont 1 % de FineFrench-v1 + corpus interne (juridique, tech, conversation)
Entraînement	24 × H100 80 Go · 500 k steps · AdamW · LR cosine
Licence	Apache-2.0

4 • Exemple d’utilisation (base)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"
model_id = "legmlai/legml-v1.0-base"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto"
)

prompt = "Donne un résumé concis de la Déclaration des Droits de l’Homme et du Citoyen de 1789."
inputs = tokenizer(prompt, return_tensors="pt").to(device)

outputs = model.generate(
    **inputs,
    temperature=0.45,
    top_p=0.9,
    max_new_tokens=256
)

print(tokenizer.decode(outputs[0, inputs['input_ids'].shape[-1]:], skip_special_tokens=True))

5 • Exemple d’utilisation (instruct & chat)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "legmlai/legml-v1.0-instruct"
device = "cuda" if torch.cuda.is_available() else "cpu"

tok  = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto"
)

messages = [
    {"role": "system", "content": "Tu es un assistant francophone rigoureux et bienveillant."},
    {"role": "user",   "content": "Explique-moi la relativité restreinte en trois points."}
]

prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tok(prompt, return_tensors="pt").to(device)

out = model.generate(
    **inputs,
    temperature=0.4,
    top_p=0.9,
    max_new_tokens=512,
    repetition_penalty=1.05
)

print(tok.decode(out[0, inputs['input_ids'].shape[-1]:], skip_special_tokens=True))

6 • Hyper-paramètres conseillés

Cas d’usage	Température	top-p	max_new_tokens
Réponse courte & factuelle	0.30-0.50	0.90	128-256
Explication détaillée	0.40-0.60	0.90	512-768
Création littéraire	0.70-0.90	0.95	≥ 512

7 • Limitations

Couverture post-avril 2025 : connaissances partielles — vérifiez les faits récents.
Maths olympiades : raisonnement complexe encore perfectible.
Biais : malgré le focus francophone, certains biais des sources persistent.

8 • Citation

@misc{legml2025,
  title  = {legml-v1.0 base & instruct: French LLMs},
  author = {Mohamad Alhajar},
  howpublished = {\url{https://huggingface.co/legmlai/legml-v1.0-base}},
  year   = {2025}
}