Create app.py

app.py

@@ -0,0 +1,170 @@
+import gradio as gr
+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
+import pandas as pd
+from datetime import datetime
+import base64
+import io
+import json
+
+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-v0.2")
+    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]
+"""
+
+# Initialize an empty DataFrame to store the history
+history_df = pd.DataFrame(columns=['Timestamp', 'Request', 'Response'])
+
+def save_history():
+    history_df.to_json('chat_history.json', orient='records')
+
+def load_history():
+    global history_df
+    try:
+        history_df = pd.read_json('chat_history.json', orient='records')
+    except FileNotFoundError:
+        history_df = pd.DataFrame(columns=['Timestamp', 'Request', 'Response'])
+    return history_df
+
+def models(text, model="Mixtral 8x7B", seed=42):
+    global history_df
+    
+    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
+    
+    # Add the current interaction to the history DataFrame
+    new_row = pd.DataFrame({
+        'Timestamp': [datetime.now().strftime("%Y-%m-%d %H:%M:%S")],  # Convert to string
+        'Request': [text],
+        'Response': [output]
+    })
+    history_df = pd.concat([history_df, new_row], ignore_index=True)
+    save_history()
+    
+    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, gr.Audio.update(interactive=True)
+
+def display_history():
+    return load_history()
+
+def download_history():
+    csv_buffer = io.StringIO()
+    history_df.to_csv(csv_buffer, index=False)
+    csv_string = csv_buffer.getvalue()
+    b64 = base64.b64encode(csv_string.encode()).decode()
+    href = f'data:text/csv;base64,{b64}'
+    return gr.HTML(f'<a href="{href}" download="chat_history.csv">Download Chat History</a>')
+
+DESCRIPTION = """ # <center><b>JARVIS⚡</b></center>
+        ### <center>A personal Assistant of Tony Stark for YOU
+        ### <center>Voice Chat with your personal Assistant</center>
+        """
+
+with gr.Blocks(css="style.css") as demo:    
+    gr.Markdown(DESCRIPTION)
+    with gr.Row():
+        select = gr.Dropdown([
+            'Mixtral 8x7B',
+            'Llama 3 8B',
+            'Mistral 7B v0.3',
+            'Phi 3 mini',
+        ],
+        value="Mistral 7B v0.3",
+        label="Model"
+        )
+        seed = gr.Slider(
+            label="Seed",
+            minimum=0,
+            maximum=999999,
+            step=1,
+            value=0,
+            visible=False
+        )
+    
+    input_audio = gr.Audio(label="User", sources="microphone", type="filepath")
+    output_audio = gr.Audio(label="AI", type="filepath", autoplay=True)
+    
+    # Add a DataFrame to display the history
+    history_display = gr.DataFrame(label="Query History")
+    
+    # Add a download button for the history
+    download_button = gr.Button("Download History")
+    download_link = gr.HTML()
+    
+    def process_audio(audio, model, seed):
+        response, audio_update = asyncio.run(respond(audio, model, seed))
+        return response, audio_update, display_history()
+
+    input_audio.change(
+        fn=process_audio,
+        inputs=[input_audio, select, seed],
+        outputs=[output_audio, input_audio, history_display]
+    )
+    
+    # Connect the download button to the download function
+    download_button.click(fn=download_history, outputs=[download_link])
+
+    # Load history when the page is refreshed
+    demo.load(fn=display_history, outputs=[history_display])
+
+if __name__ == "__main__":
+    demo.queue(max_size=200).launch(share=True)