Instructions to use CapitainVigs/mece-tunepigier-9b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use CapitainVigs/mece-tunepigier-9b with Transformers:

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

pipe = pipeline("text-generation", model="CapitainVigs/mece-tunepigier-9b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("CapitainVigs/mece-tunepigier-9b")
model = AutoModelForCausalLM.from_pretrained("CapitainVigs/mece-tunepigier-9b", device_map="auto")
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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use CapitainVigs/mece-tunepigier-9b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CapitainVigs/mece-tunepigier-9b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CapitainVigs/mece-tunepigier-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CapitainVigs/mece-tunepigier-9b

SGLang

How to use CapitainVigs/mece-tunepigier-9b 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 "CapitainVigs/mece-tunepigier-9b" \
    --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": "CapitainVigs/mece-tunepigier-9b",
		"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 "CapitainVigs/mece-tunepigier-9b" \
        --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": "CapitainVigs/mece-tunepigier-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use CapitainVigs/mece-tunepigier-9b with Docker Model Runner:
```
docker model run hf.co/CapitainVigs/mece-tunepigier-9b
```

MƐCE — Tune_Pigier (Qwen 3.5 9B fine-tuné Fongbe, modèle fusionné)

Modèle de langue conversationnel centré sur le Fongbe (fon), cœur de l'assistant vocal MƐCE. Version fusionnée (poids QLoRA mergés dans le modèle de base), chargeable directement avec transformers.

Détails

Base : Qwen 3.5 9B (multilingue), tokenizer étendu de +82 tokens Fongbe (caractères ɔ, ɛ, ɖ + sous-mots) → byte-fallback ramené de 49,24 % à 0 %.
Méthode : QLoRA (NF4, r=64, α=128, dropout 0,05), embeddings dégelées (embed_tokens, lm_head), calcul bf16. ~116 M paramètres entraînables (1,28 %).
Entraînement : 3 époques, 2× RTX 5090 (DDP), ~140 K exemples (cf. dataset associé). Loss eval finale 0,378, mean token accuracy 90,0 %.
Capacités : conversation en Fongbe, français préservé, traduction FR/EN↔FON.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

tok = AutoTokenizer.from_pretrained("CapitaineVigs/mece-tunepigier-9b")
model = AutoModelForCausalLM.from_pretrained(
    "CapitaineVigs/mece-tunepigier-9b", torch_dtype=torch.bfloat16, device_map="auto")

messages = [
    {"role": "system", "content": "Hwi wɛ nyí MƐ CE, alɔgɔ́nútɔ́ gbɛtɔ́ lɛ tɔn. A nɔ na xósin ɖò Fɔngbè mɛ hwebǐnu."},
    {"role": "user", "content": "A fɔn ganji a ?"},
]
prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
out = model.generate(**tok(prompt, return_tensors="pt").to(model.device), max_new_tokens=256)
print(tok.decode(out[0], skip_special_tokens=True))

Note d'inférence : forcer l'EOS sur <|im_end|> et bannir les single-byte tokens non-ASCII évite les artefacts (séquelle corrigée en phase 4c). Cf. app/llm.py du projet.

Limites

STT/TTS non inclus (modèles tiers intégrés dans l'app). Corpus encore en partie biblique~; pas de benchmark conversationnel Fongbe standardisé.

Licence

à reporter explicitement ici.

Citation

HOUNSOU Vignikin Gerès, MƐCE : assistant vocal Fongbe basé sur un grand modèle de langue fine-tuné, École PIGIER Bénin, 2026.

Downloads last month: 19

Safetensors

Model size

9B params

Tensor type

BF16