File size: 1,004 Bytes
abae4ba
cef7739
42a9e51
cef7739
55e69e1
 
 
 
 
 
7fa7ea5
55e69e1
abae4ba
55e69e1
 
42a9e51
 
8c7c0d7
cef7739
 
55e69e1
abae4ba
55e69e1
abae4ba
 
 
55e69e1
 
 
abae4ba
 
 
 
55e69e1
8c7c0d7
abae4ba
ab1672e
abae4ba
55e69e1
abae4ba
cef7739
42a9e51
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
from llama_cpp import Llama
import gradio as gr
import os

# 📦 Téléchargement simple (stable)
MODEL_URL = "https://huggingface.co/Pi3141/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"
MODEL_PATH = "tinyllama.gguf"

if not os.path.exists(MODEL_PATH):
    os.system(f"wget -O {MODEL_PATH} {MODEL_URL}")

# 🧠 Charger le modèle (optimisé free tier)
llm = Llama(
    model_path=MODEL_PATH,
    n_ctx=256,
    n_threads=os.cpu_count(),
    n_batch=128,
    verbose=False
)

# 💬 Chat rapide + français forcé
def chat(user_input):
    prompt = f"Réponds en français.\nQuestion: {user_input}\nRéponse:"

    output = llm(
        prompt,
        max_tokens=40,
        temperature=0.3,
        stop=["Question:"]
    )

    return output["choices"][0]["text"].strip()

# 🌐 Interface web
demo = gr.Interface(
    fn=chat,
    inputs=gr.Textbox(lines=2, placeholder="Pose ta question..."),
    outputs="text",
    title="Mon IA (TinyLlama FR)"
)

demo.launch()