Update README.md

README.md

@@ -1,58 +1,115 @@
----
-base_model: NousResearch/Meta-Llama-3-8B
-library_name: transformers
-model_name: LLama3
-tags:
-- generated_from_trainer
-- trl
-- sft
-licence: license
----
+# Llama 3 Domain Name Generator (LoRA fine-tuned)
 
-# Model Card for LLama3
+Ce modèle est une version fine-tunée de [NousResearch/Meta-Llama-3-8B](https://huggingface.co/NousResearch/Meta-Llama-3-8B) pour la génération de noms de domaine disponibles pour des entreprises.
 
-This model is a fine-tuned version of [NousResearch/Meta-Llama-3-8B](https://huggingface.co/NousResearch/Meta-Llama-3-8B).
-It has been trained using [TRL](https://github.com/huggingface/trl).
+---
 
-## Quick start
+## 🚀 **Utilisation rapide**
 
 ```python
-from transformers import pipeline
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+from peft import PeftModel
+import torch
+
+# Identifiants du modèle
+peft_model_id = "Thehunter99/LLama3"
+base_model_id = "NousResearch/Meta-Llama-3-8B"
+
+# Device
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Tokenizer
+tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+tokenizer.pad_token = tokenizer.eos_token
 
-question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
-generator = pipeline("text-generation", model="Thehunter99/LLama3", device="cuda")
-output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
-print(output["generated_text"])
+# Config QLoRA (optionnel)
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.bfloat16 if device == "cuda" else torch.float32,
+    bnb_4bit_use_double_quant=True,
+)
+
+# Charger le modèle de base
+base_model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    quantization_config=bnb_config if device == "cuda" else None,
+    device_map="auto" if device == "cuda" else None,
+    torch_dtype=torch.bfloat16 if device == "cuda" else torch.float32
+)
+
+# Charger l'adapter LoRA
+model = PeftModel.from_pretrained(base_model, peft_model_id)
+model.eval()
+if device == "cpu":
+    model = model.to("cpu")
+
+# Fonction d'inférence
+def generate_domains(prompt, max_new_tokens=50, temperature=0.7):
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            do_sample=True,
+            temperature=temperature
+        )
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Exemple de prompt (à respecter strictement)
+prompt = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+Generate available domain names for businesses. Use only .com, .io, .app, .co TLDs.
+Output format: comma-separated domains<|eot_id|>
+<|start_header_id|>user<|end_header_id|>
+organic bakery in berlin<|eot_id|>
+<|start_header_id|>assistant<|end_header_id|>"""
+
+result = generate_domains(prompt, max_new_tokens=30, temperature=0.1)
+print(result)
 ```
 
-## Training procedure
+---
 
- 
+## 📝 **Format du prompt d’inférence**
 
+**Respectez exactement ce format :**
+```
+<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+Generate available domain names for businesses. Use only .com, .io, .app, .co TLDs.
+Output format: comma-separated domains<|eot_id|>
+<|start_header_id|>user<|end_header_id|>
+[description de l'entreprise]<|eot_id|>
+<|start_header_id|>assistant<|end_header_id|>
+```
+
+**Exemple :**
+```
+<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+Generate available domain names for businesses. Use only .com, .io, .app, .co TLDs.
+Output format: comma-separated domains<|eot_id|>
+<|start_header_id|>user<|end_header_id|>
+eco-friendly coffee shop in Paris<|eot_id|>
+<|start_header_id|>assistant<|end_header_id|>
+```
 
-This model was trained with SFT.
+---
 
-### Framework versions
+## 📋 **Conseils**
 
-- TRL: 0.20.0
-- Transformers: 4.54.1
-- Pytorch: 2.6.0+cu124
-- Datasets: 4.0.0
-- Tokenizers: 0.21.2
+- Utilisez le même prompt que ci-dessus pour de meilleurs résultats.
+- Le modèle retourne une liste de domaines séparés par des virgules.
+- Pour des résultats reproductibles, utilisez `temperature=0.1`.
 
-## Citations
+---
+
+## 📚 **Dataset d’entraînement**
+
+Le modèle a été fine-tuné sur un dataset synthétique de descriptions d’entreprises et de suggestions de domaines.
+
+---
 
+## ❓ **Questions**
 
+Pour toute question ou suggestion, ouvrez une issue sur le repo ou contactez-moi sur Hugging Face.
 
-Cite TRL as:
-    
-```bibtex
-@misc{vonwerra2022trl,
-	title        = {{TRL: Transformer Reinforcement Learning}},
-	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
-	year         = 2020,
-	journal      = {GitHub repository},
-	publisher    = {GitHub},
-	howpublished = {\url{https://github.com/huggingface/trl}}
-}
-```
+---