Update app.py

app.py

@@ -5,20 +5,29 @@ import gradio as gr
 import openai
 
 print(os.environ)
-openai.api_base1 = os.environ.get("OPENAI_API_BASE")
+openai.api_base1 = os.environ.get("OPENAI_API_BASE1")
 openai.api_base2 = os.environ.get("OPENAI_API_BASE2")
-openai.api_key = os.environ.get("OPENAI_API_KEY")
+openai.api_key1 = os.environ.get("OPENAI_API_KEY")
+openai.api_key2 = os.environ.get("OPENAI_API_KEY")
+openai.api_model1 = os.environ.get("OPENAI_API_MODEL1")
+openai.api_model2 = os.environ.get("OPENAI_API_MODEL2")
 
 BASE_SYSTEM_MESSAGE = """"""
 
 
-def make_prediction1(prompt, max_tokens=None, temperature=None, top_p=None, top_k=None, repetition_penalty=None):
-    completion = openai.Completion.create(api_base=openai.api_base1, model="wizardcoder-python-34b-v1.0.Q5_K_M.gguf", prompt=prompt, max_tokens=max_tokens, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty, stream=True, stop=["</s>", "<|im_end|>"])
-    for chunk in completion:
-        yield chunk["choices"][0]["text"]
-
-def make_prediction2(prompt, max_tokens=None, temperature=None, top_p=None, top_k=None, repetition_penalty=None):
-    completion = openai.Completion.create(api_base=openai.api_base2, model="wizardcoder-python-34b-v1.0.Q3_K_M.gguf", prompt=prompt, max_tokens=max_tokens, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty, stream=True, stop=["</s>", "<|im_end|>"])
+def make_prediction(prompt, model, max_tokens=None, temperature=None, top_p=None, top_k=None, repetition_penalty=None, api_key, api_base):
+    completion = openai.Completion.create(
+        model=model,
+        api_key=api_key,
+        api_base=api_base,
+        prompt=prompt,
+        max_tokens=max_tokens,
+        temperature=temperature,
+        top_p=top_p,
+        top_k=top_k,
+        repetition_penalty=repetition_penalty,
+        stream=True,
+        stop=["</s>", "<|im_end|>"])
     for chunk in completion:
         yield chunk["choices"][0]["text"]
 
@@ -36,38 +45,14 @@ def user(message, history):
     return message, history
 
 
-def chat1(history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
-    history = history or []
-
-    messages = BASE_SYSTEM_MESSAGE + system_message.strip() + "\n" + \
-               "\n".join(["\n".join(["###Instruction\n"+item[0]+"\n\n", "###Response\n"+item[1]+"\n\n"])
-                          for item in history])
-    # strip the last `<|end_of_turn|>` from the messages
-    #messages = messages.rstrip("<|end_of_turn|>")
-    # remove last space from assistant, some models output a ZWSP if you leave a space
-    messages = messages.rstrip()
+def user_double(message, history1, history2):
+    history1 = history1 or []
+    history2 = history2 or []
+    history1.append([message, ""])
+    history2.append([message, ""])
+    return "", history1, history2
 
-    prediction = make_prediction1(
-        messages,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        top_p=top_p,
-        top_k=top_k,
-        repetition_penalty=repetition_penalty,
-    )
-    for tokens in prediction:
-        tokens = re.findall(r'(.*?)(\s|$)', tokens)
-        for subtoken in tokens:
-            subtoken = "".join(subtoken)
-            # Remove "Response\n" if it's at the beginning of the assistant's output
-            if subtoken.startswith("Response"):
-                subtoken = subtoken[len("Response"):]
-            answer = subtoken
-            history[-1][1] += answer
-            # stream the response
-            yield history, history, ""
-
-def chat2(history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
+def chat(model, history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
     history = history or []
 
     messages = BASE_SYSTEM_MESSAGE + system_message.strip() + "\n" + \
@@ -78,8 +63,9 @@ def chat2(history, system_message, max_tokens, temperature, top_p, top_k, repeti
     # remove last space from assistant, some models output a ZWSP if you leave a space
     messages = messages.rstrip()
 
-    prediction = make_prediction2(
+    prediction = make_prediction(
         messages,
+        model,
         max_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
@@ -98,6 +84,11 @@ def chat2(history, system_message, max_tokens, temperature, top_p, top_k, repeti
             # stream the response
             yield history, history, ""
 
+def chat_double(history1, history2, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
+    gen1 = chat(openai.api_model1, history1, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty, openai.api_key1, openai.api_base1)
+    gen2 = chat(openai.api_model2, history2, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty, openai.api_key2, openai.api_base2)
+    for h1, h2 in zip(gen1, gen2):
+        yield h1, h2
 
 start_message = ""
 
@@ -110,7 +101,6 @@ CSS ="""
 #chatbot2 { flex-grow: 1; overflow: auto; resize: vertical; }
 """
 
-#with gr.Blocks() as demo:
 with gr.Blocks(css=CSS) as demo:
     with gr.Row():
         with gr.Column():
@@ -122,9 +112,9 @@ with gr.Blocks(css=CSS) as demo:
     with gr.Row():
         with gr.Column():
             #chatbot = gr.Chatbot().style(height=500)
-            chatbot1 = gr.Chatbot(label="Chatbot1", elem_id="chatbot1")
+            chatbot1 = gr.Chatbot(label="Chat1: "+openai.api_model1, elem_id="chatbot1")
         with gr.Column():
-            chatbot2 = gr.Chatbot(label="Chatbot2", elem_id="chatbot2")
+            chatbot2 = gr.Chatbot(label="Chat2: "+openai.api_model2, elem_id="chatbot2")
     with gr.Row():
         message = gr.Textbox(
             label="What do you want to chat about?",
@@ -145,7 +135,7 @@ with gr.Blocks(css=CSS) as demo:
                 repetition_penalty = gr.Slider(0.0, 2.0, label="Repetition Penalty", step=0.05, value=1.1)
 
         system_msg = gr.Textbox(
-            start_message, label="System Message", interactive=True, visible=True, placeholder="System prompt. Provide instructions which you want the model to remember.", lines=5)
+            start_message, label="System Message", interactive=True, visible=True, placeholder="System prompt. Provide instructions which you want the model to remember.", lines=3)
 
     chat_history_state1 = gr.State()
     chat_history_state2 = gr.State()
@@ -154,17 +144,13 @@ with gr.Blocks(css=CSS) as demo:
     clear.click(lambda: None, None, chatbot1, queue=False)
     clear.click(lambda: None, None, chatbot2, queue=False)
 
-    submit_click_event1 = submit.click(
-        fn=user, inputs=[message, chat_history_state1], outputs=[message, chat_history_state1], queue=True
-    ).then(
-        fn=chat1, inputs=[chat_history_state1, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[chatbot1, chat_history_state1, message], queue=True
-    )
-    submit_click_event2 = submit.click(
-        fn=user, inputs=[message, chat_history_state2], outputs=[message, chat_history_state2], queue=True
+    submit_click_event = submit.click(
+        fn=user_double, inputs=[message, chat_history_state1, chat_history_state2], outputs=[message, chat_history_state1, chat_history_state2], queue=True
     ).then(
-        fn=chat2, inputs=[chat_history_state2, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[chatbot1, chat_history_state2, message], queue=True
+        fn=chat_double, inputs=[chat_history_state1, chat_history_state2, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[chatbot1, chatbot2, chat_history_state1, chat_history_state2, message], queue=True
     )
-    stop.click(fn=None, inputs=None, outputs=None, cancels=[submit_click_event1, submit_click_event2], queue=False)
+
+    stop.click(fn=None, inputs=None, outputs=None, cancels=[submit_click_event], queue=False)
 
 demo.queue(max_size=48, concurrency_count=8).launch(debug=True, server_name="0.0.0.0", server_port=7860)