Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,40 +1,36 @@
|
|
| 1 |
-
import gradio as gr
|
| 2 |
-
import torch
|
| 3 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 4 |
import os
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
-
# 🔹
|
| 7 |
-
# Le token sera automatiquement chargé dans les variables d'environnement
|
| 8 |
hf_token = os.environ.get("HF_TOKEN")
|
| 9 |
-
|
| 10 |
-
# Vérification si le token est bien chargé
|
| 11 |
if hf_token is None:
|
| 12 |
-
raise ValueError("⚠️ Le
|
| 13 |
-
"Vérifie que tu l’as bien ajouté dans les secrets du Space.")
|
| 14 |
|
| 15 |
-
# 🔹 Charger le tokenizer
|
| 16 |
-
tokenizer =
|
| 17 |
"SafaaAI/final_llm_darija_fr_tech",
|
| 18 |
-
|
|
|
|
| 19 |
trust_remote_code=True
|
| 20 |
)
|
| 21 |
|
| 22 |
# 🔹 Charger le modèle
|
| 23 |
model = AutoModelForCausalLM.from_pretrained(
|
| 24 |
"SafaaAI/final_llm_darija_fr_tech",
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
|
|
|
| 28 |
)
|
| 29 |
|
| 30 |
-
# 🔹 Fonction d
|
| 31 |
def chat_with_model(message, history):
|
| 32 |
history = history or []
|
| 33 |
-
|
| 34 |
-
# Encode texte
|
| 35 |
inputs = tokenizer(message, return_tensors="pt").to(model.device)
|
| 36 |
|
| 37 |
-
# Générer la réponse
|
| 38 |
with torch.no_grad():
|
| 39 |
output_ids = model.generate(
|
| 40 |
**inputs,
|
|
@@ -44,14 +40,13 @@ def chat_with_model(message, history):
|
|
| 44 |
temperature=0.7
|
| 45 |
)
|
| 46 |
|
| 47 |
-
# Décoder la sortie
|
| 48 |
response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
|
| 49 |
history.append((message, response))
|
| 50 |
return history, history
|
| 51 |
|
| 52 |
# 🔹 Interface Gradio
|
| 53 |
with gr.Blocks() as demo:
|
| 54 |
-
gr.Markdown("## 💬 Chatbot SafaaAI - LLM
|
| 55 |
|
| 56 |
chatbot = gr.Chatbot()
|
| 57 |
msg = gr.Textbox(label="Écris ton message ici")
|
|
@@ -62,6 +57,6 @@ with gr.Blocks() as demo:
|
|
| 62 |
msg.submit(chat_with_model, [msg, state], [chatbot, state])
|
| 63 |
clear.click(lambda: ([], []), None, [chatbot, state])
|
| 64 |
|
| 65 |
-
# 🔹 Lancer l
|
| 66 |
if __name__ == "__main__":
|
| 67 |
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
+
import torch
|
| 3 |
+
import gradio as gr
|
| 4 |
+
from transformers import RobertaTokenizerFast, AutoModelForCausalLM
|
| 5 |
|
| 6 |
+
# 🔹 Récupération du token Hugging Face depuis les "Secrets" du Space
|
|
|
|
| 7 |
hf_token = os.environ.get("HF_TOKEN")
|
|
|
|
|
|
|
| 8 |
if hf_token is None:
|
| 9 |
+
raise ValueError("⚠️ Le secret HF_TOKEN n’est pas défini dans ton Space.")
|
|
|
|
| 10 |
|
| 11 |
+
# 🔹 Charger le tokenizer (en ignorant tokenizer.json corrompu)
|
| 12 |
+
tokenizer = RobertaTokenizerFast.from_pretrained(
|
| 13 |
"SafaaAI/final_llm_darija_fr_tech",
|
| 14 |
+
use_fast=True,
|
| 15 |
+
token=hf_token,
|
| 16 |
trust_remote_code=True
|
| 17 |
)
|
| 18 |
|
| 19 |
# 🔹 Charger le modèle
|
| 20 |
model = AutoModelForCausalLM.from_pretrained(
|
| 21 |
"SafaaAI/final_llm_darija_fr_tech",
|
| 22 |
+
device_map="auto", # GPU si dispo, sinon CPU
|
| 23 |
+
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
| 24 |
+
token=hf_token,
|
| 25 |
+
trust_remote_code=True
|
| 26 |
)
|
| 27 |
|
| 28 |
+
# 🔹 Fonction d’inférence (chatbot)
|
| 29 |
def chat_with_model(message, history):
|
| 30 |
history = history or []
|
| 31 |
+
|
|
|
|
| 32 |
inputs = tokenizer(message, return_tensors="pt").to(model.device)
|
| 33 |
|
|
|
|
| 34 |
with torch.no_grad():
|
| 35 |
output_ids = model.generate(
|
| 36 |
**inputs,
|
|
|
|
| 40 |
temperature=0.7
|
| 41 |
)
|
| 42 |
|
|
|
|
| 43 |
response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
|
| 44 |
history.append((message, response))
|
| 45 |
return history, history
|
| 46 |
|
| 47 |
# 🔹 Interface Gradio
|
| 48 |
with gr.Blocks() as demo:
|
| 49 |
+
gr.Markdown("## 💬 Chatbot SafaaAI - LLM Darija + Français + Technique")
|
| 50 |
|
| 51 |
chatbot = gr.Chatbot()
|
| 52 |
msg = gr.Textbox(label="Écris ton message ici")
|
|
|
|
| 57 |
msg.submit(chat_with_model, [msg, state], [chatbot, state])
|
| 58 |
clear.click(lambda: ([], []), None, [chatbot, state])
|
| 59 |
|
| 60 |
+
# 🔹 Lancer l’app
|
| 61 |
if __name__ == "__main__":
|
| 62 |
demo.launch()
|