import gradio as gr
import requests
import json
import os
API_TOKEN = os.getenv("HF_API_TOKEN")
TRANSCRIBE_API_URL = "https://api-inference.huggingface.co/models/openai/whisper-base.en"
LLM_API_URL = "https://api-inference.huggingface.co/models/mistralai/Mixtral-8x7B-Instruct-v0.1"
def transcribe_audio(audio_file):
"""Transcribe audio file to text."""
headers = {"Authorization": f"Bearer {API_TOKEN}"}
with open(audio_file, "rb") as f:
data = f.read()
response = requests.post(TRANSCRIBE_API_URL, headers=headers, data=data)
transcription = json.loads(response.content.decode("utf-8")).get("text", "Transcription not available")
return transcription
def get_answer(context, question):
"""Get an answer from the LLM based on the context and question."""
prompt = (
"As an intelligent coding assistant, your task is to provide clear, concise, and accurate answers to coding-related questions. "
"Below are examples of questions and the kind of direct answers expected:\n\n"
"Example Question 1: How can I remove duplicates from a list in Python?\n"
"Example Answer 1: Use the set() function to convert the list to a set, which removes duplicates, then convert it back to a list.\n\n"
"Example Question 2: What's the difference between '==' and '===' in JavaScript?\n"
"Example Answer 2: '==' checks for equality of values after type coercion, while '===' checks for both value and type equality without coercion.\n\n"
"Example Question 3: How to check if a key exists in a dictionary in Python?\n"
"Example Answer 3: Use the 'in' keyword, like 'if key in my_dict:'.\n\n"
"Based on the above examples, answer the following question:\n\n"
f"Question: {question}\n"
"Answer:"
)
headers = {"Authorization": f"Bearer {API_TOKEN}"}
# Adjust generation parameters for more focused and relevant responses
payload = {
"inputs": prompt,
"parameters": {
"temperature": 0.3, # More deterministic
"top_p": 0.95, # Consider top 90% probable tokens at each step
"repetition_penalty": 1.2, # Discourage repetition
"num_return_sequences": 1, # Number of responses to generate
"return_full_text": False, # Return only generated text, not the full prompt
"top_k" : 50,
"truncate" : 24576,
"max_new_tokens" : 8192,
"stop" : [""]
},
"options": {
"use_cache": True # Use cached responses when available
}
}
response = requests.post(LLM_API_URL, headers=headers, json=payload)
answer = json.loads(response.content.decode("utf-8"))[0].get("generated_text", "Answer not available")
return answer
def transcribe_and_answer(audio_file, question):
"""Process the audio file for transcription and use the result to get an answer to a question."""
transcription = transcribe_audio(audio_file)
answer = get_answer(transcription, question)
return transcription, answer
# Create the Gradio app
import gradio as gr
# Create the Gradio app
with gr.Blocks() as app:
gr.HTML("""
TALKING DUCK
An Audio to Text Q&A Chatbot
Your swift coding sidekick. Speak your code queries, and let the duck do the magic.
""")
gr.Markdown("""
Models running on backend
""")
with gr.Row():
audio_input = gr.Audio(type="filepath", label="Upload your audio question")
question_input = gr.Textbox(label="Type your question here")
answer_button = gr.Button("Get Answer")
with gr.Row():
transcription_output = gr.Textbox(label="Transcription")
answer_output = gr.Textbox(label="Answer")
answer_button.click(transcribe_and_answer, inputs=[audio_input, question_input], outputs=[transcription_output, answer_output])
if __name__ == "__main__":
app.launch(server_name="0.0.0.0")
