Add GPT-4 Turbo

README.md

@@ -24,7 +24,7 @@ You can deploy your own bot, or use mine: [@chatgpt_karfly_bot](https://t.me/cha
 - Low latency replies (it usually takes about 3-5 seconds)
 - No request limits
 - Message streaming (watch demo)
-- GPT-4 support
+- GPT-4 and GPT-4 Turbo support
 - Group Chat support (/help_group_chat to get instructions)
 - DALLE 2 (choose 👩‍🎨 Artist mode to generate images)
 - Voice message recognition
@@ -62,6 +62,7 @@ If you want to add payments to your bot and create profitable business – write
 - *8 Mar 2023*: Added voice message recognition with [OpenAI Whisper API](https://openai.com/blog/introducing-chatgpt-and-whisper-apis). Record a voice message and ChatGPT will answer you!
 - *2 Mar 2023*: Added support of [ChatGPT API](https://platform.openai.com/docs/guides/chat/introduction).
 - *1 Aug 2023*: Added OpenAI API Base to config (useful while using OpenAI-compatible API like [LocalAI](https://github.com/go-skynet/LocalAI))
+- *15 Nov 2023*: Added support of [GPT-4 Turbo](https://help.openai.com/en/articles/8555510-gpt-4-turbo)
 
 ## Bot commands
 - `/retry` – Regenerate last bot answer

bot/openai_utils.py

@@ -22,7 +22,7 @@ OPENAI_COMPLETION_OPTIONS = {
 
 class ChatGPT:
     def __init__(self, model="gpt-3.5-turbo"):
-        assert model in {"text-davinci-003", "gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"}, f"Unknown model: {model}"
+        assert model in {"text-davinci-003", "gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4", "gpt-4-1106-preview"}, f"Unknown model: {model}"
         self.model = model
 
     async def send_message(self, message, dialog_messages=[], chat_mode="assistant"):
@@ -33,7 +33,7 @@ class ChatGPT:
         answer = None
         while answer is None:
             try:
-                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"}:
+                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4", "gpt-4-1106-preview"}:
                     messages = self._generate_prompt_messages(message, dialog_messages, chat_mode)
                     r = await openai.ChatCompletion.acreate(
                         model=self.model,
@@ -73,7 +73,7 @@ class ChatGPT:
         answer = None
         while answer is None:
             try:
-                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"}:
+                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4", "gpt-4-1106-preview"}:
                     messages = self._generate_prompt_messages(message, dialog_messages, chat_mode)
                     r_gen = await openai.ChatCompletion.acreate(
                         model=self.model,
@@ -161,6 +161,9 @@ class ChatGPT:
         elif model == "gpt-4":
             tokens_per_message = 3
             tokens_per_name = 1
+        elif model == "gpt-4-1106-preview":
+            tokens_per_message = 3
+            tokens_per_name = 1
         else:
             raise ValueError(f"Unknown model: {model}")
 

config/models.yml

@@ -1,4 +1,4 @@
-available_text_models: ["gpt-3.5-turbo", "gpt-3.5-turbo-16k", "gpt-4", "text-davinci-003"]
+available_text_models: ["gpt-3.5-turbo", "gpt-3.5-turbo-16k", "gpt-4-1106-preview", "gpt-4", "text-davinci-003"]
 
 info:
   gpt-3.5-turbo:
@@ -40,6 +40,19 @@ info:
       Fast: 2
       Cheap: 2
 
+  gpt-4-1106-preview:
+    type: chat_completion
+    name: GPT-4 Turbo
+    description: GPT-4 Turbo is a <b>faster</b> and <b>cheaper</b> version of GPT-4. It's as smart as GPT-4, so you should use it instead of GPT-4.
+
+    price_per_1000_input_tokens: 0.01
+    price_per_1000_output_tokens: 0.03
+
+    scores:
+      smart: 5
+      fast: 4
+      cheap: 3
+
   text-davinci-003:
     type: completion
     name: GPT-3.5