Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -9,17 +9,13 @@ import spaces
|
|
| 9 |
import torch
|
| 10 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
| 11 |
|
| 12 |
-
# Description de la démo
|
| 13 |
DESCRIPTION = "# Mistral-7B v0.3"
|
| 14 |
-
|
| 15 |
-
# Si pas de GPU détecté, afficher un message
|
| 16 |
if not torch.cuda.is_available():
|
| 17 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
| 18 |
|
| 19 |
-
MAX_MAX_NEW_TOKENS = 2048
|
| 20 |
-
DEFAULT_MAX_NEW_TOKENS = 1024
|
| 21 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
| 22 |
|
|
|
|
| 23 |
if torch.cuda.is_available():
|
| 24 |
model_id = "mistralai/Mistral-7B-Instruct-v0.3"
|
| 25 |
model = AutoModelForCausalLM.from_pretrained(
|
|
@@ -33,7 +29,6 @@ if torch.cuda.is_available():
|
|
| 33 |
def generate(
|
| 34 |
message: str,
|
| 35 |
chat_history: list[dict],
|
| 36 |
-
# Valeurs par défaut pour la génération
|
| 37 |
max_new_tokens: int = 1024,
|
| 38 |
temperature: float = 0.6,
|
| 39 |
top_p: float = 0.9,
|
|
@@ -41,22 +36,19 @@ def generate(
|
|
| 41 |
repetition_penalty: float = 1.2,
|
| 42 |
) -> Iterator[str]:
|
| 43 |
"""
|
| 44 |
-
Génération de texte à partir de l'historique
|
| 45 |
"""
|
| 46 |
conversation = [*chat_history, {"role": "user", "content": message}]
|
| 47 |
-
|
| 48 |
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
|
|
|
|
|
|
| 49 |
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
| 50 |
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
| 51 |
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
|
|
|
| 52 |
input_ids = input_ids.to(model.device)
|
| 53 |
|
| 54 |
-
streamer = TextIteratorStreamer(
|
| 55 |
-
tokenizer,
|
| 56 |
-
timeout=20.0,
|
| 57 |
-
skip_prompt=True,
|
| 58 |
-
skip_special_tokens=True
|
| 59 |
-
)
|
| 60 |
generate_kwargs = dict(
|
| 61 |
{"input_ids": input_ids},
|
| 62 |
streamer=streamer,
|
|
@@ -76,37 +68,131 @@ def generate(
|
|
| 76 |
outputs.append(text)
|
| 77 |
yield "".join(outputs)
|
| 78 |
|
| 79 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
custom_css = """
|
| 81 |
html, body {
|
| 82 |
height: 100%;
|
| 83 |
margin: 0;
|
| 84 |
padding: 0;
|
| 85 |
background: linear-gradient(135deg, #FDE2E2, #E2ECFD) !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
}
|
| 87 |
"""
|
| 88 |
|
| 89 |
-
#
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
|
| 111 |
if __name__ == "__main__":
|
| 112 |
demo.queue(max_size=20).launch()
|
|
|
|
| 9 |
import torch
|
| 10 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
| 11 |
|
|
|
|
| 12 |
DESCRIPTION = "# Mistral-7B v0.3"
|
|
|
|
|
|
|
| 13 |
if not torch.cuda.is_available():
|
| 14 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
| 15 |
|
|
|
|
|
|
|
| 16 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
| 17 |
|
| 18 |
+
# Chargement du modèle (sur GPU si disponible)
|
| 19 |
if torch.cuda.is_available():
|
| 20 |
model_id = "mistralai/Mistral-7B-Instruct-v0.3"
|
| 21 |
model = AutoModelForCausalLM.from_pretrained(
|
|
|
|
| 29 |
def generate(
|
| 30 |
message: str,
|
| 31 |
chat_history: list[dict],
|
|
|
|
| 32 |
max_new_tokens: int = 1024,
|
| 33 |
temperature: float = 0.6,
|
| 34 |
top_p: float = 0.9,
|
|
|
|
| 36 |
repetition_penalty: float = 1.2,
|
| 37 |
) -> Iterator[str]:
|
| 38 |
"""
|
| 39 |
+
Génération de texte à partir de l'historique + message utilisateur.
|
| 40 |
"""
|
| 41 |
conversation = [*chat_history, {"role": "user", "content": message}]
|
|
|
|
| 42 |
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
| 43 |
+
|
| 44 |
+
# Gestion de la limite de tokens
|
| 45 |
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
| 46 |
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
| 47 |
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
| 48 |
+
|
| 49 |
input_ids = input_ids.to(model.device)
|
| 50 |
|
| 51 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
generate_kwargs = dict(
|
| 53 |
{"input_ids": input_ids},
|
| 54 |
streamer=streamer,
|
|
|
|
| 68 |
outputs.append(text)
|
| 69 |
yield "".join(outputs)
|
| 70 |
|
| 71 |
+
# Liste des questions (chaque question est associée à une "bulle cliquable")
|
| 72 |
+
QUESTIONS = [
|
| 73 |
+
"I - C’est quoi le consentement ? Comment savoir si ma copine a envie de moi ?",
|
| 74 |
+
"II - C’est quoi une agression sexuelle ?",
|
| 75 |
+
"III - C’est quoi un viol ?",
|
| 76 |
+
"IV - C’est quoi un attouchement ?",
|
| 77 |
+
"V - C’est quoi un harcèlement sexuel ?",
|
| 78 |
+
"VI - Est ce illégal de visionner du porno ?",
|
| 79 |
+
"VII - Mon copain me demande un nude, dois-je le faire ?",
|
| 80 |
+
"VIII - Mon ancien copain me menace de poster des photos de moi nue sur internet, que faire ?",
|
| 81 |
+
"IX - Que puis-je faire si un membre de ma famille me touche d’une manière bizarre, mais que j’ai peur de parler ou de ne pas être cru ?",
|
| 82 |
+
]
|
| 83 |
+
|
| 84 |
+
# -- CSS personnalisé --
|
| 85 |
custom_css = """
|
| 86 |
html, body {
|
| 87 |
height: 100%;
|
| 88 |
margin: 0;
|
| 89 |
padding: 0;
|
| 90 |
background: linear-gradient(135deg, #FDE2E2, #E2ECFD) !important;
|
| 91 |
+
font-family: sans-serif;
|
| 92 |
+
}
|
| 93 |
+
|
| 94 |
+
/* Conteneur global pour centrer la "pieuvre" */
|
| 95 |
+
.octopus-container {
|
| 96 |
+
position: relative;
|
| 97 |
+
width: 600px;
|
| 98 |
+
height: 600px;
|
| 99 |
+
margin: 0 auto; /* centre horizontalement */
|
| 100 |
+
margin-top: 30px; /* marge en haut */
|
| 101 |
+
}
|
| 102 |
+
|
| 103 |
+
/* L'image (point d'interrogation) au centre */
|
| 104 |
+
.center-image {
|
| 105 |
+
position: absolute;
|
| 106 |
+
width: 100px; /* Ajuster la taille */
|
| 107 |
+
top: 50%;
|
| 108 |
+
left: 50%;
|
| 109 |
+
transform: translate(-50%, -50%);
|
| 110 |
+
}
|
| 111 |
+
|
| 112 |
+
/* Bulles de question */
|
| 113 |
+
.octopus-arm {
|
| 114 |
+
position: absolute;
|
| 115 |
+
width: 140px;
|
| 116 |
+
background: #FFFFFFCC;
|
| 117 |
+
padding: 8px;
|
| 118 |
+
border-radius: 10px;
|
| 119 |
+
box-shadow: 0 2px 4px rgba(0,0,0,0.2);
|
| 120 |
+
text-align: center;
|
| 121 |
+
cursor: pointer; /* Pour montrer que c'est cliquable */
|
| 122 |
+
}
|
| 123 |
+
|
| 124 |
+
/* Positions "rayonnantes" */
|
| 125 |
+
.arm1 { top: 0%; left: 50%; transform: translate(-50%, 0); }
|
| 126 |
+
.arm2 { top: 10%; left: 80%; }
|
| 127 |
+
.arm3 { top: 35%; left: 95%; transform: translateX(-100%); }
|
| 128 |
+
.arm4 { top: 60%; left: 80%; }
|
| 129 |
+
.arm5 { top: 85%; left: 50%; transform: translate(-50%, -100%); }
|
| 130 |
+
.arm6 { top: 60%; left: 10%; }
|
| 131 |
+
.arm7 { top: 35%; left: 0%; }
|
| 132 |
+
.arm8 { top: 10%; left: 10%; }
|
| 133 |
+
.arm9 { top: 0%; left: 25%; }
|
| 134 |
+
|
| 135 |
+
/* Texte dans les bulles */
|
| 136 |
+
.octopus-arm p {
|
| 137 |
+
margin: 0;
|
| 138 |
+
font-size: 0.85rem;
|
| 139 |
+
line-height: 1.2;
|
| 140 |
+
color: #000;
|
| 141 |
}
|
| 142 |
"""
|
| 143 |
|
| 144 |
+
#
|
| 145 |
+
# Construction de l'application avec Blocks
|
| 146 |
+
#
|
| 147 |
+
with gr.Blocks(css=custom_css) as demo:
|
| 148 |
+
# Titre / Description
|
| 149 |
+
gr.Markdown(DESCRIPTION)
|
| 150 |
+
|
| 151 |
+
# Section HTML : la "pieuvre" + le script JS
|
| 152 |
+
# Notez la fonction setChatText(...) qu'on appelle en cliquant sur la bulle.
|
| 153 |
+
with gr.Box():
|
| 154 |
+
gr.HTML(
|
| 155 |
+
"""
|
| 156 |
+
<div class="octopus-container">
|
| 157 |
+
<!-- Image point d'interrogation au centre -->
|
| 158 |
+
<img src="https://cdn-icons-png.flaticon.com/512/4926/4926733.png" class="center-image" />
|
| 159 |
+
|
| 160 |
+
<!-- Bulles de questions (cliquables) -->
|
| 161 |
+
""" +
|
| 162 |
+
"".join(
|
| 163 |
+
f"""
|
| 164 |
+
<div class="octopus-arm arm{i+1}" onclick="setChatText(`{QUESTIONS[i]}`)">
|
| 165 |
+
<p>{QUESTIONS[i]}</p>
|
| 166 |
+
</div>
|
| 167 |
+
"""
|
| 168 |
+
for i in range(len(QUESTIONS))
|
| 169 |
+
) +
|
| 170 |
+
"""
|
| 171 |
+
</div>
|
| 172 |
|
| 173 |
+
<script>
|
| 174 |
+
// Au clic sur une bulle, insère la question dans la zone de saisie du ChatInterface
|
| 175 |
+
function setChatText(questionText) {
|
| 176 |
+
// Le ChatInterface de Gradio possède un <textarea> dont
|
| 177 |
+
// l'attribut placeholder est souvent "Type a message..."
|
| 178 |
+
// On cherche ce textarea et on y met la question.
|
| 179 |
+
let textBox = document.querySelector('textarea[placeholder="Type a message..."]');
|
| 180 |
+
if (textBox) {
|
| 181 |
+
textBox.value = questionText;
|
| 182 |
+
} else {
|
| 183 |
+
console.warn("Impossible de trouver la zone de texte du chatbot.");
|
| 184 |
+
}
|
| 185 |
+
}
|
| 186 |
+
</script>
|
| 187 |
+
"""
|
| 188 |
+
)
|
| 189 |
+
|
| 190 |
+
# Le chatbot Gradio, en dessous
|
| 191 |
+
chat = gr.ChatInterface(
|
| 192 |
+
fn=generate,
|
| 193 |
+
type="messages",
|
| 194 |
+
description="Cliquez sur une question ci-dessus ou saisissez votre texte.",
|
| 195 |
+
)
|
| 196 |
|
| 197 |
if __name__ == "__main__":
|
| 198 |
demo.queue(max_size=20).launch()
|