Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,17 +3,14 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
|
| 3 |
|
| 4 |
MODEL = "prithivMLmods/Llama-SmolTalk-3.2-1B-Instruct"
|
| 5 |
|
| 6 |
-
# Charger le tokenizer
|
| 7 |
tokenizer = AutoTokenizer.from_pretrained(MODEL)
|
| 8 |
|
| 9 |
-
# Charger le modèle en 8 bits
|
| 10 |
model = AutoModelForCausalLM.from_pretrained(
|
| 11 |
MODEL,
|
| 12 |
device_map="auto",
|
| 13 |
load_in_8bit=True
|
| 14 |
)
|
| 15 |
|
| 16 |
-
# Pipeline
|
| 17 |
chatbot = pipeline(
|
| 18 |
"text-generation",
|
| 19 |
model=model,
|
|
@@ -21,41 +18,42 @@ chatbot = pipeline(
|
|
| 21 |
device_map="auto"
|
| 22 |
)
|
| 23 |
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
"et donne uniquement ta réponse."
|
| 30 |
-
)
|
| 31 |
|
| 32 |
-
def chat(message, history):
|
| 33 |
-
history = history
|
| 34 |
-
|
| 35 |
-
|
|
|
|
| 36 |
|
| 37 |
resp = chatbot(
|
| 38 |
prompt,
|
| 39 |
-
max_new_tokens=
|
| 40 |
do_sample=True,
|
| 41 |
temperature=0.7,
|
| 42 |
top_p=0.9,
|
| 43 |
repetition_penalty=1.1
|
| 44 |
)[0]["generated_text"]
|
| 45 |
|
| 46 |
-
#
|
| 47 |
-
reply = resp.split("
|
| 48 |
-
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
-
history.append(
|
|
|
|
|
|
|
| 51 |
return history, history
|
| 52 |
|
| 53 |
with gr.Blocks() as demo:
|
| 54 |
-
|
| 55 |
-
state = gr.State([]) # sauvegarde de l'historique
|
| 56 |
msg = gr.Textbox(placeholder="Écris un message...")
|
| 57 |
-
|
| 58 |
-
msg.submit(chat, [msg, state], [chatbot_ui, state])
|
| 59 |
-
msg.submit(lambda: "", None, msg) # reset input après envoi
|
| 60 |
|
| 61 |
demo.launch()
|
|
|
|
| 3 |
|
| 4 |
MODEL = "prithivMLmods/Llama-SmolTalk-3.2-1B-Instruct"
|
| 5 |
|
|
|
|
| 6 |
tokenizer = AutoTokenizer.from_pretrained(MODEL)
|
| 7 |
|
|
|
|
| 8 |
model = AutoModelForCausalLM.from_pretrained(
|
| 9 |
MODEL,
|
| 10 |
device_map="auto",
|
| 11 |
load_in_8bit=True
|
| 12 |
)
|
| 13 |
|
|
|
|
| 14 |
chatbot = pipeline(
|
| 15 |
"text-generation",
|
| 16 |
model=model,
|
|
|
|
| 18 |
device_map="auto"
|
| 19 |
)
|
| 20 |
|
| 21 |
+
system_prompt = """Tu es Aria, une IA gentille, claire et polie.
|
| 22 |
+
Réponds toujours en phrases complètes.
|
| 23 |
+
Ne fais pas de jeu de rôle.
|
| 24 |
+
Ne répète pas le message de l'utilisateur.
|
| 25 |
+
Réponds directement à la question."""
|
|
|
|
|
|
|
| 26 |
|
| 27 |
+
def chat(message, history=[]):
|
| 28 |
+
history = history[-3:] # garder 3 derniers échanges
|
| 29 |
+
|
| 30 |
+
context = "\n".join([f"Utilisateur: {m[0]}\nAria: {m[1]}" for m in history])
|
| 31 |
+
prompt = f"{system_prompt}\n{context}\nUtilisateur: {message}\nAria:"
|
| 32 |
|
| 33 |
resp = chatbot(
|
| 34 |
prompt,
|
| 35 |
+
max_new_tokens=120, # plus long
|
| 36 |
do_sample=True,
|
| 37 |
temperature=0.7,
|
| 38 |
top_p=0.9,
|
| 39 |
repetition_penalty=1.1
|
| 40 |
)[0]["generated_text"]
|
| 41 |
|
| 42 |
+
# Garde seulement après "Aria:"
|
| 43 |
+
reply = resp.split("Aria:")[-1].strip()
|
| 44 |
+
|
| 45 |
+
# Supprime si "Utilisateur:" est revenu dans la génération
|
| 46 |
+
if "Utilisateur:" in reply:
|
| 47 |
+
reply = reply.split("Utilisateur:")[0].strip()
|
| 48 |
|
| 49 |
+
history.append([message, reply])
|
| 50 |
+
|
| 51 |
+
# Retourner au format Gradio Chatbot
|
| 52 |
return history, history
|
| 53 |
|
| 54 |
with gr.Blocks() as demo:
|
| 55 |
+
chat_ui = gr.Chatbot()
|
|
|
|
| 56 |
msg = gr.Textbox(placeholder="Écris un message...")
|
| 57 |
+
msg.submit(chat, [msg, chat_ui], [chat_ui, chat_ui])
|
|
|
|
|
|
|
| 58 |
|
| 59 |
demo.launch()
|