| import streamlit as st |
| import edge_tts |
| import asyncio |
| import tempfile |
| import os |
| from huggingface_hub import InferenceClient |
| import re |
| from streaming_stt_nemo import Model |
| import torch |
| import random |
|
|
| default_lang = "en" |
|
|
| engines = {default_lang: Model(default_lang)} |
|
|
| def transcribe(audio): |
| lang = "en" |
| model = engines[lang] |
| text = model.stt_file(audio)[0] |
| return text |
|
|
| HF_TOKEN = os.environ.get("HF_TOKEN", None) |
|
|
| def client_fn(model): |
| if "Mixtral" in model: |
| return InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1") |
| elif "Llama" in model: |
| return InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct") |
| elif "Mistral" in model: |
| return InferenceClient("mistralai/Mistral-7B-Instruct-v0.3") |
| elif "Phi" in model: |
| return InferenceClient("microsoft/Phi-3-mini-4k-instruct") |
| else: |
| return InferenceClient("microsoft/Phi-3-mini-4k-instruct") |
|
|
| def randomize_seed_fn(seed: int) -> int: |
| seed = random.randint(0, 999999) |
| return seed |
|
|
| system_instructions1 = """ |
| [SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark.' |
| Keep conversation friendly, short, clear, and concise. |
| Avoid unnecessary introductions and answer the user's questions directly. |
| Respond in a normal, conversational manner while being friendly and helpful. |
| [USER] |
| """ |
|
|
| def models(text, model="Mixtral 8x7B", seed=42): |
|
|
| seed = int(randomize_seed_fn(seed)) |
| generator = torch.Generator().manual_seed(seed) |
|
|
| client = client_fn(model) |
|
|
| generate_kwargs = dict( |
| max_new_tokens=300, |
| seed=seed |
| ) |
|
|
| formatted_prompt = system_instructions1 + text + "[JARVIS]" |
| stream = client.text_generation( |
| formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False) |
| output = "" |
| for response in stream: |
| if not response.token.text == "</s>": |
| output += response.token.text |
|
|
| return output |
|
|
| async def respond(audio, model, seed): |
| user = transcribe(audio) |
| reply = models(user, model, seed) |
| communicate = edge_tts.Communicate(reply) |
| with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: |
| tmp_path = tmp_file.name |
| await communicate.save(tmp_path) |
| return tmp_path |
|
|
| st.title("JARVIS⚡") |
| st.markdown("### A personal Assistant of Tony Stark for YOU") |
| st.markdown("### Voice Chat with your personal Assistant") |
|
|
| with st.form("voice_form"): |
| model_choice = st.selectbox("Choose a model", ['Mixtral 8x7B', 'Llama 3 8B', 'Mistral 7B v0.3', 'Phi 3 mini'], key="voice_model") |
| audio_file = st.file_uploader("Upload Audio", type=["wav", "mp3"], key="voice_audio") |
| submit_button = st.form_submit_button("Submit") |
|
|
| if submit_button: |
| if audio_file is not None: |
| with st.spinner("Transcribing and generating response..."): |
| audio_bytes = audio_file.read() |
| with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: |
| tmp_file.write(audio_bytes) |
| tmp_path = tmp_file.name |
| response = respond(tmp_path, model_choice, 42) |
| st.audio(response, format='audio/wav') |
|
|
| with st.form("text_form"): |
| model_choice = st.selectbox("Choose a model", ['Mixtral 8x7B', 'Llama 3 8B', 'Mistral 7B v0.3', 'Phi 3 mini'], key="text_model") |
| user_text = st.text_area("Enter your message:", key="text_input") |
| submit_button = st.form_submit_button("Submit") |
|
|
| if submit_button: |
| if user_text: |
| with st.spinner("Generating response..."): |
| response = models(user_text, model_choice, 42) |
| st.text_area("JARVIS Response", value=response, key="text_output", height=150) |
