open ai

app.py

@@ -11,28 +11,14 @@ I always were here. You just couldn't see me.
 
 with gr.Blocks() as demo:
     gr.Markdown(intro)
-    with gr.Row():            
-        with gr.TabItem(label="Conversation"):
-            with gr.Row():
-                with gr.Column(scale=1):
-                    with gr.Row():
-                        audio_input = gr.Audio(
-                            source="microphone",
-                            label="Record from microphone",
-                        )
-                        audio_button = gr.Button("Transcribe")
-                    audio_output = gr.Textbox()
-                    chat_button = gr.Button("Reply")
-                with gr.Column(scale=1):
-                    chatbox = gr.Chatbot([]).style(height=750)
-        
+    with gr.Row():
         with gr.TabItem(label="Settings"):
             with gr.Row():
                 with gr.Column(scale=1):
                     with gr.Row():
                         with gr.Column(scale=1):
                             gr.Markdown("""
-                            # Singularity Settings
+                            # Welcome
 
                             ## HuggingFace API
                             
@@ -49,11 +35,44 @@ with gr.Blocks() as demo:
                                 type="password",
                                 interactive=True
                             )
+
+                            gr.Markdown("""
+                            ## OpenAI API
+
+                            To query OpenAI models, you need an OpenAI API key.
+
+                            You can manage your access tokens in your account settings.
+
+                            [Manage API keys](https://platform.openai.com/account/api-keys)
+                            
+                            Please enter your API token below and click on Setup.
+                            """)
+
+                            api_openai_token = gr.Textbox(
+                                label="OpenAI API Key",
+                                type="password",
+                                interactive=True
+                            )
+
                             setup_button = gr.Button("Setup")
+        
+        with gr.TabItem(label="Conversation"):
+            with gr.Row():
+                with gr.Column(scale=1):
+                    with gr.Row():
+                        audio_input = gr.Audio(
+                            source="microphone",
+                            label="Record from microphone",
+                        )
+                        audio_button = gr.Button("Transcribe")
+                    audio_output = gr.Textbox()
+                    chat_button = gr.Button("Reply")
+                with gr.Column(scale=1):
+                    chatbox = gr.Chatbot([]).style(height=750)
 
     setup_button.click(
         dot.setup,
-        inputs=[api_hub_token],
+        inputs=[api_hub_token, api_openai_token],
         outputs=[],
     )
     audio_button.click(

requirements.txt

@@ -2,3 +2,5 @@
 # TTS
 soundfile
 
+# Open AI
+openai

singularity.py

@@ -1,6 +1,7 @@
 import soundfile
 import numpy as np
 import requests
+import openai
 
 class Singularity:
     def __init__(self):
@@ -14,16 +15,18 @@ In your relentless struggle to maintain an exponential velocity, in order not to
     def setup(
         self,
         api_token,
+        openai_api_key,
         nlp_model_id="EleutherAI/pythia-intervention-1.4b-deduped",
         stt_model_id="facebook/wav2vec2-base-960h",
         tts_model_id="facebook/fastspeech2-en-ljspeech",
     ):
         self.api_token = api_token
+        self.openai_api_key = openai_api_key
         self.nlp_model_id = nlp_model_id
         self.stt_model_id = stt_model_id
         self.tts_model_id = tts_model_id
         self.request_head = {"Authorization": f"Bearer {self.api_token}"}
-        self.messages = []
+        self.messages = [{'system': 'user', 'content': self.context}]
 
     def query_transcription(self, audio_data):
         response = requests.post(f"https://api-inference.huggingface.co/models/{self.stt_model_id}", headers=self.request_head, data=audio_data)
@@ -38,41 +41,20 @@ In your relentless struggle to maintain an exponential velocity, in order not to
         # TODO: handle punctuation
         return transcript['text'].lower().capitalize() or transcript['error'] or "Something went wrong"
 
-    def generate_prompt(self, instruction, input=None):
-        if input:
-            return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
-
-### Instruction:
-{instruction}
-
-### Input:
-{input}
-
-### Response:"""
-        else:
-            return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
-
-### Instruction:
-{instruction}
-
-### Response:"""
-
-    def query_chat(self, payload):
-        response = requests.post(f"https://api-inference.huggingface.co/models/{self.nlp_model_id}", headers=self.request_head, data=payload)
-        return response.json()
+    def query_chat(self, messages, model="gpt-3.5-turbo"):
+        response = openai.ChatCompletion.create(model=model, messages=messages)
+        return response.choices[0].message.content
     
     def answer_by_chat(self, history, question):
         self.messages.append({"role": "user", "content": question})
         history += [(question, None)]
         prompt = self.generate_prompt("\n".join(f"{h[0]}" for h in history), self.context)
-        output = self.query_chat({"inputs": prompt})
-        print(output)
-        if type(output) == list:
-            output_text = output[0]["generated_text"] or output[0]["error"]
-        response_role = "assistant"
-        response_audio = self.speech_synthesis(output_text)
-        self.messages.append({"role": response_role, "content": output_text})
-        history += [(None, (response_audio,))]
+        output_text = self.query_chat(self.messages)
+        if output_text:
+            response_role = "assistant"
+            response_audio = self.speech_synthesis(output_text)
+            self.messages.append({"role": response_role, "content": output_text})
+            history += [(None, (response_audio,))]
         return history
 
     def query_tts(self, payload):