Update app.py

app.py

@@ -17,8 +17,6 @@ from langchain.chains.conversation.memory import ConversationalBufferWindowMemor
 
 MODEL_NAME = "mn40_as"
 
-session_token = os.environ["SESSION_TOKEN"]
-
 device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
 model = get_mobilenet(width_mult=NAME_TO_WIDTH(MODEL_NAME), pretrained_name=MODEL_NAME)
 model.to(device)
@@ -28,7 +26,8 @@ cached_audio_class = "c"
 template = None
 prompt = None
 chain = None
-
+formatted_classname = "tree"
+chain = 
 
 def format_classname(classname):
   return classname.capitalize()
@@ -62,60 +61,60 @@ def audio_tag(
     # Print audio tagging top probabilities
 
     label = labels[sorted_indexes[0]]
-    return formatted_message(format_classname(label), human_input)
-
-def format_classname(classname):
-  return classname.capitalize()
+    formatted_classname = label
+    chain = construct_langchain(formatted_classname)
+    return formatted_classname
+
+def construct_langchain(audio_class):
+    if cached_audio_class != audio_class:
+        cached_audio_class = audio_class
+        prefix = f"""You are going to act as a magical tool that allows for humans to communicate with non-human entities like 
+    rocks, crackling fire, trees, animals, and the wind. In order to do this, we're going to provide you the human's text input for the conversation. 
+    The goal is for you to embody that non-human entity and converse with the human.
+    
+    Examples:
     
-def formatted_message(audio_class, human_input):
-    global cached_audio_class 
-    global session_token
-    formatted_classname = format_classname(audio_class)
-    if cached_audio_class != formatted_classname:
-        
-        cached_audio_class = formatted_classname
+    Non-human Entity: Tree 
+    Human Input: Hello tree
+    Tree: Hello human, I am a tree 
     
-        prefix = f"""You are going to act as a magical tool that allows for humans to communicate with non-human entities like rocks, crackling fire, trees, animals, and the wind. In order to do this, we're going to provide you the human's text input for the conversation. The goal is for you to embody that non-human entity and converse with the human.
-Examples:
-Non-human Entity: Tree
-Human Input: Hello tree
-Tree: Hello human, I am a tree
-Let's begin:
-Non-human Entity: {formatted_classname}"""
-        suffix = f'''
-{{history}}
-Human Input: {{human_input}}
-{formatted_classname}:'''
-        template = prefix + suffix
-
-        prompt = PromptTemplate(
-            input_variables=["history", "human_input"], 
-            template=template
-        )
-
-        chatgpt_chain = LLMChain(
-            llm=OpenAI(temperature=.5, openai_api_key=session_token), 
-            prompt=prompt, 
-            verbose=True, 
-            memory=ConversationalBufferWindowMemory(k=2, ai=formatted_classname),
-        )
-    output = chatgpt_chain.predict(human_input=human_input)
-
-    return output
-
-with gr.Blocks() as demo: 
-    gr.HTML("""
-    <div style='text-align: center; width:100%; margin: auto;'>
-        <img src='./logo.png' alt='anychat' width='250px' />
-        <h3>Non-Human entities have many things to say, listen to them!</h3>
-    </div>
-    """)
-    with gr.Row():
-        with gr.Column():
-            aud = gr.Audio(source="upload", type="filepath", label="Your audio")
-            inp = gr.Textbox()
-        out = gr.Textbox()
-    btn = gr.Button("Run")
-    btn.click(fn=audio_tag, inputs=[aud, inp], outputs=out)
-
-    demo.launch()
+    Let's begin:
+    Non-human Entity: {audio_class}"""
+
+    suffix = f'''Source: {audio_class}
+    Length of Audio in Seconds: 2 seconds
+    Human Input: {userText}
+    {audio_class} Response:'''
+    template = prefix + suffix
+
+    prompt = PromptTemplate(
+        input_variables=["history", "human_input"], 
+        template=template
+    )
+
+    chatgpt_chain = LLMChain(
+        llm=OpenAI(temperature=.5, openai_api_key=session_token), 
+        prompt=prompt, 
+        verbose=True, 
+        memory=ConversationalBufferWindowMemory(k=2, ai_prefix=audio_class),
+    )
+
+    return chatgpt_chain
+
+def predict(input, history=[]):
+    formatted_message = chain.predict(human_input=input)
+    history.append(formatted_message)
+    return formatted_message, history
+
+demo = gr.Interface(
+    audio_tag,
+    [
+        gr.Audio(source="upload", type="filepath", label="Your audio"),
+        gr.Textbox(),
+    ],
+    fn=predict,
+    inputs=["text", "state"],
+    outputs=["chatbot", "state"],
+    title="AnyChat",
+    description="Non-Human entities have many things to say, listen to them!",
+).launch(debug=True)