import time import openai import gradio as gr import requests from pydub import AudioSegment as am from xml.etree import ElementTree aoai_url, aoai_key, stts_key, stts_region = "", "", "", "" openai.api_type = "azure" prompts = "" model_gpt = "" messages_gpt = [] model_chat = "" messages_chat = [ {"role": "system", "content": "You are an AI assistant that helps people find information."}, ] response_walle = [] model_vchat = "" messages_vchat = [ {"role": "system", "content": "You are an AI assistant that helps people find information and just respond with SSML."}, ] def get_aoai_set(get_aoai_url, get_aoai_key, get_aoai_API): if get_aoai_url: openai.api_base = get_aoai_url if get_aoai_key: openai.api_key = get_aoai_key if get_aoai_API: openai.api_version = get_aoai_API return gr.update(value=get_aoai_url), gr.update(value=get_aoai_key), gr.update(value=get_aoai_API) def get_stts_set(get_stts_key, get_stts_region): global stts_key, stts_region if get_stts_key: stts_key = get_stts_key if get_stts_region: stts_region = get_stts_region return gr.update(value=get_stts_key), gr.update(value=get_stts_region) with gr.Blocks() as page: with gr.Tabs(): with gr.TabItem("Settings"): gr.HTML("""

Please read and set parameters before switching to another tab.


Your Azure OpenAI Key and other Azure Cognitive Service Keys will not be saved or viewed by anyone.

You can find these parameters in Azure Portal. Select Azure OpenAI resource or Cognitive Service resource like Speech, and then select 'Keys and Endpoint' from left panel.
For Azure OpenAI service, you need to provide the resource URL and key for REST API. You also need to set the API version or just use the default value. The Azure OpenAI model which is deployed needs to be set in each tab. Because you may need to run different models at the same time. Don't forget to hit 'Enter' with every input.
For Azure Cognitive services, you need to provide a Key for REST API, and also need to provide a service region, for example, 'westus'. The app will create the endpoint URL by itself.

Thank you.


Azure OpenAI Service parameters for ChatGPT/GPT. Please input these settings and hit the 'Enter' key. """) with gr.Row(): with gr.Column(scale=0.6): ui_aoai_url = gr.Textbox(placeholder="Like https://your-url-base.openai.azure.com , etc.", label="- Azure OpenAI service API endpoint:", lines=1).style(container=False) with gr.Column(scale=0.2): ui_aoai_key = gr.Textbox(placeholder="Please enter your Azure OpenAI API key here.", label="- Azure OpenAI service API Key: ", lines=1, type='password').style(container=False) with gr.Column(scale=0.2): ui_aoai_api = gr.Textbox(value="2023-03-15-preview", label="· Azure OpenAI service API version: ", lines=1, interactive=True).style(container=False) gr.HTML("Azure Cognitive Speech Service parameters to use VoiceChat. ") with gr.Row(): with gr.Column(scale=0.6): ui_stts_key = gr.Textbox(placeholder="Please enter your speech service API key if you want to try VoiceChat. " + "Please input these settings and hit 'Enter' key.", label="- Azure Cognitive Speech service API Key: ", interactive=True, type='password').style(container=False) with gr.Column(scale=0.4): ui_stts_loc = gr.Textbox(placeholder="Please enter your speech service region.", label="- Azure Cognitive Speech service region: ", interactive=True).style(container=False) ui_aoai_url.submit(get_aoai_set, [ui_aoai_url, ui_aoai_key, ui_aoai_api], [ui_aoai_url, ui_aoai_key, ui_aoai_api]) ui_aoai_key.submit(get_aoai_set, [ui_aoai_url, ui_aoai_key, ui_aoai_api], [ui_aoai_url, ui_aoai_key, ui_aoai_api]) ui_aoai_api.submit(get_aoai_set, [ui_aoai_url, ui_aoai_key, ui_aoai_api], [ui_aoai_url, ui_aoai_key, ui_aoai_api]) ui_stts_key.submit(get_stts_set, [ui_stts_key, ui_stts_loc], [ui_stts_key, ui_stts_loc]) ui_stts_loc.submit(get_stts_set, [ui_stts_key, ui_stts_loc], [ui_stts_key, ui_stts_loc]) with gr.TabItem("GPT-3.5 Playground"): ui_chatbot_gpt = gr.Chatbot(label="GPT Playground:") with gr.Row(): with gr.Column(scale=0.9): ui_prompt_gpt = gr.Textbox(placeholder="Please enter your prompt here.", show_label=False).style(container=False) with gr.Column(scale=0.1, min_width=100): ui_clear_gpt = gr.Button("Clear Input", ) with gr.Accordion("Expand to config parameters:", open=True): ui_memo_gpt = gr.HTML("GPT-3.5 playground use Completion(). So you just need to provide model name as engine parameter.") ui_model_gpt = gr.Textbox(placeholder="Azure OpenAI GPT model deployment name. ", label="- Azure OpenAI deployment name:", lines=1).style(container=False) with gr.Row(): ui_temp_gpt = gr.Slider(0.1, 1.0, 0.9, step=0.1, label="Temperature", interactive=True) ui_max_tokens_gpt = gr.Slider(100, 4000, 1000, step=100, label="Max Tokens", interactive=True) ui_top_p_gpt = gr.Slider(0.1, 1.0, 0.5, step=0.1, label="Top P", interactive=True) with gr.Accordion("Select radio button to see detail:", open=False): ui_res_radio_gpt = gr.Radio(["Response from OpenAI Model", "Prompt messages history"], label="Show OpenAI response:", interactive=True) ui_response_gpt = gr.TextArea(show_label=False, interactive=False).style(container=False) def get_parameters_gpt(slider_1, slider_2, slider_3): ui_temp_gpt.value = slider_1 ui_max_tokens_gpt.value = slider_2 ui_top_p_gpt.value = slider_3 print("Log - Updated GPT parameters: Temperature=", ui_temp_gpt.value, " Max Tokens=", ui_max_tokens_gpt.value, " Top_P=", ui_top_p_gpt.value) def get_engine_gpt(get_aoai_model): global model_gpt model_gpt = get_aoai_model return gr.update(value=get_aoai_model) def select_response_gpt(radio): if radio == "Response from OpenAI Model": return gr.update(value=gpt_x) else: return gr.update(value=messages_gpt) def user_gpt(user_message, history): global prompts prompts = user_message messages_gpt.append(prompts) return "", history + [[user_message, None]] def bot_gpt(history): global gpt_x print(ui_model_gpt.value) gpt_x = openai.Completion.create( engine=model_gpt, prompt=prompts, temperature=0.6, max_tokens=1000, top_p=1, frequency_penalty=0, presence_penalty=0, best_of=1, stop=None ) gpt_reply = gpt_x.choices[0].text messages_gpt.append(gpt_reply) history[-1][1] = gpt_reply return history ui_model_gpt.submit(get_engine_gpt, ui_model_gpt , ui_model_gpt) ui_temp_gpt.change(get_parameters_gpt, [ui_temp_gpt, ui_max_tokens_gpt, ui_top_p_gpt]) ui_max_tokens_gpt.change(get_parameters_gpt, [ui_temp_gpt, ui_max_tokens_gpt, ui_top_p_gpt]) ui_top_p_gpt.change(get_parameters_gpt, [ui_temp_gpt, ui_max_tokens_gpt, ui_top_p_gpt]) ui_prompt_gpt.submit(user_gpt, [ui_prompt_gpt, ui_chatbot_gpt], [ui_prompt_gpt, ui_chatbot_gpt], queue=False).then( bot_gpt, ui_chatbot_gpt, ui_chatbot_gpt ) ui_clear_gpt.click(lambda: None, None, ui_chatbot_gpt, queue=False) ui_res_radio_gpt.change(select_response_gpt, ui_res_radio_gpt, ui_response_gpt) with gr.TabItem("ChatGPT on GPT-4"): ui_chatbot_chat = gr.Chatbot(label="ChatGPT:") with gr.Row(): with gr.Column(scale=0.9): ui_prompt_chat = gr.Textbox(placeholder="Please enter your prompt here.", show_label=False).style(container=False) with gr.Column(scale=0.1, min_width=100): ui_clear_chat = gr.Button("Clear Chat") with gr.Blocks(): with gr.Accordion("Expand to config parameters:", open=True): gr.HTML("ChatGPT use ChatCompletion(). Here is the default system prompt, you can change it to your own prompt.") ui_prompt_sys = gr.Textbox(value="You are an AI assistant that helps people find information.", label="- Here is the default system prompt, you can change it to your own prompt.", interactive=True).style(container=False) ui_model_chat = gr.Textbox(placeholder="Azure OpenAI model deployment name. ", label="- Azure OpenAI GPT-3.5/4 deployment name:", lines=1).style(container=False) with gr.Row(): ui_temp_chat = gr.Slider(0.1, 1.0, 0.7, step=0.1, label="Temperature", interactive=True) ui_max_tokens_chat = gr.Slider(100, 8000, 2000, step=100, label="Max Tokens", interactive=True) ui_top_p_chat = gr.Slider(0.05, 1.0, 0.9, step=0.1, label="Top P", interactive=True) with gr.Accordion("Select radio button to see detail:", open=False): ui_res_radio_chat = gr.Radio(["Response from OpenAI Model", "Prompt messages history"], label="Show OpenAI response:", interactive=True) ui_response_chat = gr.TextArea(show_label=False, interactive=False).style(container=False) def get_parameters_chat(slider_1, slider_2, slider_3): ui_temp_chat.value = slider_1 ui_max_tokens_chat.value = slider_2 ui_top_p_chat.value = slider_3 print("Log - Updated chatGPT parameters: Temperature=", ui_temp_chat.value, " Max Tokens=", ui_max_tokens_chat.value, " Top_P=", ui_top_p_chat.value) def get_engine_chat(get_aoai_model): global model_chat model_chat = get_aoai_model return gr.update(value=get_aoai_model) def select_response_chat(radio): if radio == "Response from OpenAI Model": return gr.update(value=chat_x) else: return gr.update(value=messages_chat) def user_chat(user_message, history): messages_chat.append({"role": "user", "content": user_message}) return "", history + [[user_message, None]] def bot_chat(history): global chat_x chat_x = openai.ChatCompletion.create( engine=model_chat, messages=messages_chat, temperature=ui_temp_chat.value, max_tokens=ui_max_tokens_chat.value, top_p=ui_top_p_chat.value, frequency_penalty=0, presence_penalty=0, stop=None ) ui_response_chat.value= chat_x print(ui_response_chat.value) chat_reply = chat_x.choices[0].message.content messages_chat.append({"role": "assistant", "content": chat_reply}) history[-1][1] = chat_reply return history def reset_sys(sysmsg): global messages_chat messages_chat = [ {"role": "system", "content": sysmsg}, ] ui_model_chat.submit(get_engine_chat, ui_model_chat, ui_model_chat) ui_res_radio_chat.change(select_response_chat, ui_res_radio_chat, ui_response_chat) ui_temp_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_max_tokens_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_top_p_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_prompt_sys.submit(reset_sys, ui_prompt_sys) ui_prompt_chat.submit(user_chat, [ui_prompt_chat, ui_chatbot_chat], [ui_prompt_chat, ui_chatbot_chat], queue=False).then( bot_chat, ui_chatbot_chat, ui_chatbot_chat ) ui_clear_chat.click(lambda: None, None, ui_chatbot_chat, queue=False).then(reset_sys, ui_prompt_sys) with gr.TabItem("DALL·E 2 Painting"): ui_prompt_walle = gr.Textbox(placeholder="Please enter your prompt here to generate image.", show_label=False).style(container=False) ui_image_walle = gr.Image() with gr.Accordion("Select radio button to see detail:", open=False): ui_response_walle = gr.TextArea(show_label=False, interactive=False).style(container=False) def get_image_walle(prompt_walle): global response_walle walle_api_version = '2022-08-03-preview' url = "{}dalle/text-to-image?api-version={}".format(openai.api_base, walle_api_version) headers= { "api-key": openai.api_key, "Content-Type": "application/json" } body = { "caption": prompt_walle, "resolution": "1024x1024" } submission = requests.post(url, headers=headers, json=body) response_walle.append(submission.json()) print("Log - WALL·E status: {}".format(submission.json())) operation_location = submission.headers['Operation-Location'] retry_after = submission.headers['Retry-after'] status = "" while (status != "Succeeded"): time.sleep(int(retry_after)) response = requests.get(operation_location, headers=headers) response_walle.append(response.json()) print("Log - WALL·E status: {}".format(response.json())) status = response.json()['status'] image_url_walle = response.json()['result']['contentUrl'] return gr.update(value=image_url_walle) def get_response_walle(): global response_walle return gr.update(value=response_walle) ui_prompt_walle.submit(get_image_walle, ui_prompt_walle, ui_image_walle, queue=False).then(get_response_walle, None, ui_response_walle) with gr.TabItem("VoiceChat on GPT"): with gr.Row(): with gr.Column(): with gr.Accordion("Expand to config parameters:", open=True): ui_prompt_sys_vchat = gr.Textbox(value="You are an AI assistant that helps people find information and just respond with SSML.", label="- Here is the default system prompt, you can change it to your own prompt.", interactive=True).style(container=False) ui_model_vchat = gr.Textbox(placeholder="- Azure OpenAI model deployment name. ", label="- Azure OpenAI GPT-3.5/4 deployment name:", lines=1).style(container=False) ui_voice_inc_vchat = gr.Audio(source="microphone", type="filepath") ui_voice_out_vchat = gr.Audio(value=None, type="filepath", interactive=False).style(container=False) with gr.Accordion("Expand to config parameters:", open=False): with gr.Row(): ui_temp_vchat = gr.Slider(0.1, 1.0, 0.7, step=0.1, label="Temperature", interactive=True) ui_max_tokens_vchat = gr.Slider(100, 8000, 800, step=100, label="Max Tokens", interactive=True) ui_top_p_vchat = gr.Slider(0.05, 1.0, 0.9, step=0.1, label="Top P", interactive=True) with gr.Column(): ui_chatbot_vchat = gr.Chatbot(label="Voice to ChatGPT:") with gr.Accordion("Select radio button to see detail:", open=False): ui_res_radio_vchat = gr.Radio(["Response from OpenAI Model", "Prompt messages history"], label="Show OpenAI response:", interactive=True) ui_response_vchat = gr.TextArea(show_label=False, interactive=False).style(container=False) def get_parameters_vchat(slider_1, slider_2, slider_3): ui_temp_vchat.value = slider_1 ui_max_tokens_vchat.value = slider_2 ui_top_p_vchat.value = slider_3 print("Log - Updated chatGPT parameters: Temperature=", ui_temp_vchat.value, " Max Tokens=", ui_max_tokens_vchat.value, " Top_P=", ui_top_p_vchat.value) def get_engine_vchat(get_aoai_model): global model_vchat model_vchat = get_aoai_model return gr.update(value=get_aoai_model) def select_response_vchat(radio): if radio == "Response from OpenAI Model": return gr.update(value=vchat_x) else: return gr.update(value=messages_vchat) def speech_to_text(voice_message): # Downsample input voice to 16kHz voice_wav = am.from_file(voice_message, format='wav') voice_wav = voice_wav.set_frame_rate(16000) voice_wav.export(voice_message, format='wav') # STT service_region = stts_region base_url = "https://"+service_region+".stt.speech.microsoft.com/" path = 'speech/recognition/conversation/cognitiveservices/v1' constructed_url = base_url + path params = { 'language': 'zh-CN', 'format': 'detailed' } headers = { 'Ocp-Apim-Subscription-Key': stts_key, 'Content-Type': 'audio/wav; codecs=audio/pcm; samplerate=16000', 'Accept': 'application/json;text/xml' } body = open(voice_message,'rb').read() response = requests.post(constructed_url, params=params, headers=headers, data=body) if response.status_code == 200: rs = response.json() if rs != '': print(rs) else: print("\nLog - Status code: " + str(response.status_code) + "\nSomething went wrong. Check your subscription key and headers.\n") print("Reason: " + str(response.reason) + "\n") sst_text = rs['DisplayText'] return sst_text def text_to_speech(): service_region = stts_region # test print(stts_key) base_url = "https://"+service_region+".tts.speech.microsoft.com/" path = 'cognitiveservices/v1' constructed_url = base_url + path headers = { 'Ocp-Apim-Subscription-Key': stts_key, 'Content-Type': 'application/ssml+xml', 'X-Microsoft-OutputFormat': 'riff-24khz-16bit-mono-pcm', 'User-Agent': 'Voice ChatGPT' } xml_body = ElementTree.Element('speak', version='1.0') xml_body.set('{http://www.w3.org/XML/1998/namespace}lang', 'zh-cn') voice = ElementTree.SubElement(xml_body, 'voice') voice.set('{http://www.w3.org/XML/1998/namespace}lang', 'zh-cn') voice.set('name', 'zh-CN-XiaoxiaoNeural') voice.text = vchat_reply body = ElementTree.tostring(xml_body) response = requests.post(constructed_url, headers=headers, data=body) if response.status_code == 200: with open('chatgpt.wav', 'wb') as audio: audio.write(response.content) print("\nStatus code: " + str(response.status_code) + "\nYour TTS is ready for playback.\n") else: print("\nStatus code: " + str(response.status_code) + "\nSomething went wrong. Check your subscription key and headers.\n") print("Reason: " + str(response.reason) + "\n") tts_file = "chatgpt.wav" return gr.update(value=tts_file, interactive=True) def user_vchat(user_voice_message, history): user_message = speech_to_text(user_voice_message) messages_vchat.append({"role": "user", "content": user_message}) return history + [[user_message, None]] def bot_vchat(history): global vchat_x, vchat_reply vchat_x = openai.ChatCompletion.create( engine=model_vchat, messages=messages_vchat, temperature=ui_temp_chat.value, max_tokens=ui_max_tokens_chat.value, top_p=ui_top_p_chat.value, frequency_penalty=0, presence_penalty=0, stop=None ) ui_response_vchat.value= vchat_x print(ui_response_vchat.value) vchat_reply = vchat_x.choices[0].message.content messages_vchat.append({"role": "assistant", "content": vchat_reply}) history[-1][1] = vchat_reply return history ui_model_vchat.submit(get_engine_vchat, ui_model_vchat, ui_model_vchat) ui_res_radio_vchat.change(select_response_vchat, ui_res_radio_vchat, ui_response_vchat) ui_temp_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_max_tokens_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_top_p_chat.change(get_parameters_chat, [ui_temp_chat, ui_max_tokens_chat, ui_top_p_chat]) ui_voice_inc_vchat.change(user_vchat, [ui_voice_inc_vchat, ui_chatbot_vchat], ui_chatbot_vchat, queue=False).then( bot_vchat, ui_chatbot_vchat, ui_chatbot_vchat, queue=False).then(text_to_speech, None, ui_voice_out_vchat) page.launch(share=False, debug=True)