import gradio as gr import openai import fitz # PyMuPDF for PDF processing import base64 import io # Variable to store API key api_key = "" # Function to update API key def set_api_key(key): global api_key api_key = key return "API Key Set Successfully!" # Function to interact with OpenAI API def query_openai(messages, temperature, top_p, max_output_tokens): if not api_key: return "Please enter your OpenAI API key first." try: openai.api_key = api_key # Set API key dynamically # Ensure numeric values for OpenAI parameters temperature = float(temperature) if temperature else 1.0 top_p = float(top_p) if top_p else 1.0 max_output_tokens = int(max_output_tokens) if max_output_tokens else 2048 response = openai.ChatCompletion.create( model="gpt-4.1", messages=messages, temperature=temperature, top_p=top_p, max_tokens=max_output_tokens ) return response["choices"][0]["message"]["content"] except Exception as e: return f"Error: {str(e)}" # Function to process image URL input def image_url_chat(image_url, text_query, temperature, top_p, max_output_tokens): if not image_url or not text_query: return "Please provide an image URL and a query." messages = [ {"role": "user", "content": [ {"type": "image_url", "image_url": {"url": image_url}}, {"type": "text", "text": text_query} ]}, ] return query_openai(messages, temperature, top_p, max_output_tokens) # Function to process text input def text_chat(text_query, temperature, top_p, max_output_tokens): if not text_query: return "Please enter a query." messages = [{"role": "user", "content": [{"type": "text", "text": text_query}]}] return query_openai(messages, temperature, top_p, max_output_tokens) # Function to process uploaded image input def image_chat(image_file, text_query, temperature, top_p, max_output_tokens): if image_file is None or not text_query: return "Please upload an image and provide a query." # Encode image as base64 with open(image_file, "rb") as img: base64_image = base64.b64encode(img.read()).decode("utf-8") image_data = f"data:image/jpeg;base64,{base64_image}" messages = [ {"role": "user", "content": [ {"type": "image_url", "image_url": {"url": image_data}}, {"type": "text", "text": text_query} ]}, ] return query_openai(messages, temperature, top_p, max_output_tokens) # Function to process uploaded PDF input def pdf_chat(pdf_file, text_query, temperature, top_p, max_output_tokens): if pdf_file is None or not text_query: return "Please upload a PDF and provide a query." try: # Extract text from all pages of the PDF doc = fitz.open(pdf_file.name) text = "\n".join([page.get_text("text") for page in doc]) # Extract text from all pages # If no text found, return an error if not text.strip(): return "No text found in the PDF." # Create the query message with the extracted text and the user's query messages = [ {"role": "user", "content": [ {"type": "text", "text": text}, # The extracted text from the PDF {"type": "text", "text": text_query} ]}, ] return query_openai(messages, temperature, top_p, max_output_tokens) except Exception as e: return f"Error processing the PDF: {str(e)}" # Function to transcribe audio to text using OpenAI Whisper API def transcribe_audio(audio_binary, openai_api_key): if not openai_api_key: return "Error: No API key provided." openai.api_key = openai_api_key try: # Use the correct transcription API call audio_file_obj = io.BytesIO(audio_binary) audio_file_obj.name = 'audio.wav' # Set a name for the file object (as OpenAI expects it) # Transcribe the audio to text using OpenAI's whisper model audio_file_transcription = openai.Audio.transcribe(file=audio_file_obj, model="whisper-1") return audio_file_transcription.text except Exception as e: return f"Error transcribing audio: {str(e)}" # Function to handle uploaded audio transcription def process_uploaded_audio(audio_binary): if not audio_binary: return "Please upload an audio file first." if not api_key: return "Please enter your OpenAI API key first." try: transcription = transcribe_audio(audio_binary, api_key) return transcription except Exception as e: return f"Error transcribing audio: {str(e)}" # Function to handle recorded audio transcription def process_recorded_audio(audio_path): if not audio_path: return "No audio recorded." if not api_key: return "Please enter your OpenAI API key first." try: with open(audio_path, "rb") as audio_file: audio_binary = audio_file.read() transcription = transcribe_audio(audio_binary, api_key) return transcription except Exception as e: return f"Error transcribing recorded audio: {str(e)}" # Function to process the voice chat queries def process_voice_query(transcription, temperature, top_p, max_output_tokens): if not transcription or transcription.startswith("Error") or transcription.startswith("Please"): return "Please ensure audio is transcribed successfully first." # Use the transcription as the query messages = [{"role": "user", "content": [{"type": "text", "text": transcription}]}] return query_openai(messages, temperature, top_p, max_output_tokens) # Function to clear the chat - FIXED to return the correct types for file inputs def clear_chat(): # For file components like gr.File and gr.Audio, we should return None # For text components, return empty string # For sliders, return default values # The order must match exactly with the outputs in clear_button.click() return ( "", # image_url (textbox) "", # image_query (textbox) "", # image_url_output (textbox) "", # text_query (textbox) "", # text_output (textbox) "", # image_text_query (textbox) "", # image_output (textbox) None, # pdf_upload (file) "", # pdf_text_query (textbox) "", # pdf_output (textbox) None, # audio_upload (file) "", # upload_transcription (textbox) "", # upload_audio_output (textbox) None, # audio_recorder (audio) "", # record_transcription (textbox) "", # record_audio_output (textbox) 1.0, # temperature (slider) 1.0, # top_p (slider) 2048 # max_output_tokens (slider) ) # Gradio UI Layout with gr.Blocks(theme=gr.themes.Ocean()) as demo: gr.Markdown("## GPT-4.5 Preview Chatbot") with gr.Accordion("How to Use This App!", open=False, elem_id="neuroscope-accordion"): gr.Markdown(""" ### Getting Started: 1. Enter your OpenAI API key in the field at the top and click "Set API Key" 2. Adjust the hyperparameters if needed (Temperature, Top-P, Max Output Tokens) ### Using the Different Tabs: #### Image URL Chat - Paste an image URL in the field - Enter your question about the image - Click "Ask" to get a response #### Text Chat - Simply type your query in the text field - Click "Ask" to get a response #### Image Chat - Upload an image from your device - Enter your question about the uploaded image - Click "Ask" to get a response #### PDF Chat - Upload a PDF document - Ask questions about the PDF content - Click "Ask" to get a response #### Voice Chat - **Upload Audio:** Upload an audio file, click "Transcribe Audio", then click "Ask" - **Record Audio:** Record your voice, click "Transcribe Recording", then click "Ask" ### Tips: - Use the "Clear Chat" button to reset all fields - For more creative responses, try increasing the Temperature - For longer responses, increase the Max Output Tokens """) # Accordion for explaining hyperparameters with gr.Accordion("Hyperparameters", open=False, elem_id="neuroscope-accordion"): gr.Markdown(""" ### Temperature: Controls the randomness of the model's output. A lower temperature makes the model more deterministic, while a higher temperature makes it more creative and varied. ### Top-P (Nucleus Sampling): Controls the cumulative probability distribution from which the model picks the next word. A lower value makes the model more focused and deterministic, while a higher value increases randomness. ### Max Output Tokens: Limits the number of tokens (words or subwords) the model can generate in its response. You can use this to control the length of the response. """) gr.HTML(""" """) # API Key Input with gr.Row(): api_key_input = gr.Textbox(label="Enter OpenAI API Key", type="password") api_key_button = gr.Button("Set API Key", elem_id="api_key_button") api_key_output = gr.Textbox(label="API Key Status", interactive=False) with gr.Row(): temperature = gr.Slider(0, 2, value=1.0, step=0.1, label="Temperature") top_p = gr.Slider(0, 1, value=1.0, step=0.1, label="Top-P") max_output_tokens = gr.Slider(0, 16384, value=2048, step=512, label="Max Output Tokens") with gr.Tabs(): with gr.Tab("Image URL Chat"): image_url = gr.Textbox(label="Enter Image URL") image_query = gr.Textbox(label="Ask about the Image") image_url_output = gr.Textbox(label="Response", interactive=False) image_url_button = gr.Button("Ask", elem_id="ask_button") with gr.Tab("Text Chat"): text_query = gr.Textbox(label="Enter your query") text_output = gr.Textbox(label="Response", interactive=False) text_button = gr.Button("Ask", elem_id="ask_button") with gr.Tab("Image Chat"): image_upload = gr.File(label="Upload an Image", type="filepath") image_text_query = gr.Textbox(label="Ask about the uploaded image") image_output = gr.Textbox(label="Response", interactive=False) image_button = gr.Button("Ask", elem_id="ask_button") with gr.Tab("PDF Chat"): pdf_upload = gr.File(label="Upload a PDF", type="filepath") pdf_text_query = gr.Textbox(label="Ask about the uploaded PDF") pdf_output = gr.Textbox(label="Response", interactive=False) pdf_button = gr.Button("Ask", elem_id="ask_button") with gr.Tab("Voice Chat"): with gr.Tabs(): with gr.Tab("Upload Audio"): # Upload audio section audio_upload = gr.File(label="Upload an Audio File", type="binary") upload_transcribe_button = gr.Button("Transcribe Audio", elem_id="transcribe_button") upload_transcription = gr.Textbox(label="Transcription", interactive=False) upload_audio_output = gr.Textbox(label="Response", interactive=False) upload_audio_button = gr.Button("Ask", elem_id="ask_button") with gr.Tab("Record Audio"): # Record audio section audio_recorder = gr.Audio(label="Record your voice", type="filepath") record_transcribe_button = gr.Button("Transcribe Recording", elem_id="transcribe_button") record_transcription = gr.Textbox(label="Transcription", interactive=False) record_audio_output = gr.Textbox(label="Response", interactive=False) record_audio_button = gr.Button("Ask", elem_id="ask_button") # Clear chat button clear_button = gr.Button("Clear Chat", elem_id="clear_chat_button") # Button Click Actions api_key_button.click(set_api_key, inputs=[api_key_input], outputs=[api_key_output]) image_url_button.click(image_url_chat, [image_url, image_query, temperature, top_p, max_output_tokens], image_url_output) text_button.click(text_chat, [text_query, temperature, top_p, max_output_tokens], text_output) image_button.click(image_chat, [image_upload, image_text_query, temperature, top_p, max_output_tokens], image_output) pdf_button.click(pdf_chat, [pdf_upload, pdf_text_query, temperature, top_p, max_output_tokens], pdf_output) # Voice Chat - Upload Audio tab actions upload_transcribe_button.click( process_uploaded_audio, inputs=[audio_upload], outputs=[upload_transcription] ) # FIXED: Properly order the inputs to process_voice_query upload_audio_button.click( process_voice_query, inputs=[upload_transcription, temperature, top_p, max_output_tokens], outputs=[upload_audio_output] ) # Voice Chat - Record Audio tab actions record_transcribe_button.click( process_recorded_audio, inputs=[audio_recorder], outputs=[record_transcription] ) # FIXED: Properly order the inputs to process_voice_query record_audio_button.click( process_voice_query, inputs=[record_transcription, temperature, top_p, max_output_tokens], outputs=[record_audio_output] ) # Clear button resets all necessary fields clear_button.click( clear_chat, outputs=[ image_url, image_query, image_url_output, text_query, text_output, image_text_query, image_output, pdf_upload, pdf_text_query, pdf_output, audio_upload, upload_transcription, upload_audio_output, audio_recorder, record_transcription, record_audio_output, temperature, top_p, max_output_tokens ] ) # Launch Gradio App if __name__ == "__main__": demo.launch()