|
import gradio as gr |
|
from googletrans import Translator |
|
from transformers import pipeline |
|
import tempfile |
|
import edge_tts |
|
import inflect |
|
|
|
translator = Translator() |
|
number_to_word = inflect.engine() |
|
|
|
|
|
model = pipeline("text-generation", model="dbddv01/gpt2-french-small") |
|
|
|
async def transcribe(audio): |
|
lang = "fr" |
|
|
|
text = transcribe(audio) |
|
return text |
|
|
|
async def respond(audio): |
|
user = await transcribe(audio) |
|
|
|
user_fr = translator.translate(user, src='en', dest='fr').text |
|
reply = model(user_fr, max_length=50, num_return_sequences=1)[0]['generated_text'] |
|
|
|
reply_en = translator.translate(reply, src='fr', dest='en').text |
|
communicate = edge_tts.Communicate(reply_en) |
|
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: |
|
tmp_path = tmp_file.name |
|
await communicate.save(tmp_path) |
|
yield tmp_path |
|
|
|
DESCRIPTION = """ # <center><b>JARVIS⚡</b></center> |
|
### <center>A personal Assistant of Tony Stark for YOU |
|
### <center>Voice Chat with your personal Assistant</center> |
|
""" |
|
|
|
MORE = """ ## TRY Other Models |
|
### Instant Video: Create Amazing Videos in 5 Second -> https://huggingface.co/spaces/KingNish/Instant-Video |
|
### Instant Image: 4k images in 5 Second -> https://huggingface.co/spaces/KingNish/Instant-Image |
|
""" |
|
|
|
FAST = """## Fastest Model""" |
|
|
|
base_loaded = "mistralai/Mixtral-8x7B-Instruct-v0.1" |
|
|
|
client1 = InferenceClient(base_loaded) |
|
|
|
system_instructions1 = "<s>[SYSTEM] Répondez comme le vrai Jarvis JARVIS, fabriqué par 'Tony Stark'. La conversation doit être très courte, claire, amicale et concise. Le texte fourni est une demande de réponse spécifique de votre part, en tant qu'assistant virtuel. Il vous demande de fournir des réponses amicales comme si vous étiez le personnage Jarvis, créé par « Tony Stark ». On s'attend à ce que j'évite les présentations et que je réponde directement à la requête, en développant tous les aspects de la demande. En tant qu'assistant doté d'une IA, votre tâche consiste à générer des réponses qui semblent avoir été créées par l'utilisateur et non par un assistant doté d'une IA.[USER]" |
|
|
|
async def generate1(prompt): |
|
generate_kwargs = dict( |
|
temperature=0.7, |
|
max_new_tokens=512, |
|
top_p=0.95, |
|
repetition_penalty=1, |
|
do_sample=False, |
|
) |
|
formatted_prompt = system_instructions1 + prompt + "[JARVIS]" |
|
stream = client1.text_generation( |
|
formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True) |
|
output = "" |
|
for response in stream: |
|
if not response.token.text == "</s>": |
|
output += response.token.text |
|
|
|
communicate = edge_tts.Communicate(output) |
|
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: |
|
tmp_path = tmp_file.name |
|
await communicate.save(tmp_path) |
|
yield tmp_path |
|
|
|
with gr.Blocks(css="style.css") as demo: |
|
gr.Markdown(DESCRIPTION) |
|
with gr.Row(): |
|
user_input = gr.Audio(label="Voice Chat (BETA)", type="filepath") |
|
output_audio = gr.Audio(label="JARVIS", type="filepath", |
|
interactive=False, |
|
autoplay=True, |
|
elem_classes="audio") |
|
with gr.Row(): |
|
translate_btn = gr.Button("Response") |
|
translate_btn.click(fn=respond, inputs=user_input, |
|
outputs=output_audio, api_name=False) |
|
gr.Markdown(FAST) |
|
with gr.Row(): |
|
user_input = gr.Textbox(label="Prompt", value="What is Wikipedia") |
|
input_text = gr.Textbox(label="Input Text", elem_id="important") |
|
output_audio = gr.Audio(label="JARVIS", type="filepath", |
|
interactive=False, |
|
autoplay=True, |
|
elem_classes="audio") |
|
with gr.Row(): |
|
translate_btn = gr.Button("Response") |
|
translate_btn.click(fn=generate1, inputs=user_input, |
|
outputs=output_audio, api_name="translate") |
|
|
|
gr.Markdown(MORE) |
|
|
|
if __name__ == "__main__": |
|
demo.queue(max_size=200).launch() |
|
|
