Instructions to use Finisha-F-scratch/Lorias-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Finisha-F-scratch/Lorias-v1 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Finisha-F-scratch/Lorias-v1")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("Finisha-F-scratch/Lorias-v1") model = AutoModelForCausalLM.from_pretrained("Finisha-F-scratch/Lorias-v1", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Finisha-F-scratch/Lorias-v1 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Finisha-F-scratch/Lorias-v1" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Finisha-F-scratch/Lorias-v1", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Finisha-F-scratch/Lorias-v1
- SGLang
How to use Finisha-F-scratch/Lorias-v1 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 "Finisha-F-scratch/Lorias-v1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Finisha-F-scratch/Lorias-v1", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "Finisha-F-scratch/Lorias-v1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Finisha-F-scratch/Lorias-v1", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Finisha-F-scratch/Lorias-v1 with Docker Model Runner:
docker model run hf.co/Finisha-F-scratch/Lorias-v1
File size: 3,330 Bytes
cc47d18 390087d abbf276 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | ---
library_name: transformers
tags:
- Pretrained
license: other
language:
- fr
pipeline_tag: text-generation
---
# 🐺 Lorias-v1 : French Language Model From Scratch

Bienvenue dans le dépôt officiel de **Lorias-v1**, un modèle de langage (LLM) de **1 milliard de paramètres**, entièrement conçu et pré-entraîné pour la langue française par **Clémence (Finisha)**.
## 📌 Présentation du Projet
Lorias-v1 représente une étape majeure dans la création de modèles souverains et spécialisés. Contrairement à de nombreux modèles qui se contentent d'un affinage (fine-tuning), Lorias-v1 a été bâti **"from scratch"**, avec une architecture optimisée et un **tokenizer propriétaire** spécifiquement entraîné sur des corpus francophones.
* **Développeuse :** Clémence (Finisha)
* **Taille :** 1 Milliard de paramètres
* **Langue :** Français 🇫🇷
* **Statut :** Base Model (Pre-trained)
---
## 🚀 Caractéristiques Techniques
### 🏗️ Architecture & Tokenisation
* **Architecture :** Transformer decoder-only optimisé.
* **Tokenizer :** Custom French Tokenizer, conçu pour capturer les nuances morphologiques de la langue française et réduire la segmentation abusive des mots complexes.
* **Vocabulaire :** Optimisé pour maximiser l'efficacité du contexte en français.
### 🧠 Capacités Actuelles
Le modèle est actuellement en phase de **post-pré-entraînement**. Ses performances se concentrent sur :
* **Maîtrise Syntaxique :** Une excellente compréhension de la structure des phrases et de la grammaire française.
* **Fluidité :** Capacité à générer des séquences de mots cohérentes sur le plan structurel.
* **Connecteurs Logiques :** Utilisation de charnières linguistiques pour lier les concepts.
> **Note importante :** En tant que modèle de base non affiné, Lorias-v1 se concentre sur la structure. La cohérence sémantique profonde et la spécialisation aux instructions seront apportées lors des prochaines phases de **Fine-tuning (SFT)**.
---
## 📊 Performances & Comportement
Actuellement, Lorias-v1 montre une propension à générer des listes terminologiques reliées par des connecteurs inattendus. Ce comportement est typique d'un modèle de 1B qui a parfaitement assimilé la **forme** de la langue avant d'en stabiliser le **fond**.
| Aspect | Statut |
| --- | --- |
| **Syntaxe** | ✅ Excellent |
| **Orthographe** | ✅ Maîtrisé |
| **Sémantique** | 🔄 En cours (via futur Fine-tuning) |
| **Logique métier** | 📅 Prévu pour v1.1 |
---
## 🛠️ Utilisation (Roadmap)
1. **Phase 1 (Actuelle) :** Exploration de la structure linguistique et complétion de texte.
2. **Phase 2 (Prochainement) :** Fine-tuning supervisé (Instruct) pour aligner les réponses aux intentions des utilisateurs.
3. **Phase 3 :** Optimisation de la sémantique et réduction des hallucinations structurelles.
---
## 🤝 Contribuer & Contact
Le projet Lorias est en constante évolution. Si vous souhaitez échanger sur l'architecture ou les données d'entraînement :
* **Développeuse :** Clémence (Finisha)
* **Objectif :** Créer un écosystème de modèles performants et accessibles pour la communauté francophone.
---
**Fait avec ❤️ par Clémence.** |