From 533e6705c839dab5b22927fef873071cf4b7f9c1 Mon Sep 17 00:00:00 2001
From: maccagnit <45689045+maccagnit@users.noreply.github.com>
Date: Tue, 20 Jun 2023 10:47:19 +0200
Subject: [PATCH] Added GPT-3.5 16K chat functionality (#306)

---
 bot/openai_utils.py | 11 +++++++----
 config/models.yml   | 17 +++++++++++++++--
 2 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/bot/openai_utils.py b/bot/openai_utils.py
index cdb0110..5a68302 100644
--- a/bot/openai_utils.py
+++ b/bot/openai_utils.py
@@ -16,7 +16,7 @@ OPENAI_COMPLETION_OPTIONS = {
 
 class ChatGPT:
     def __init__(self, model="gpt-3.5-turbo"):
-        assert model in {"text-davinci-003", "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"}, f"Unknown model: {model}"
         self.model = model
 
     async def send_message(self, message, dialog_messages=[], chat_mode="assistant"):
@@ -27,7 +27,7 @@ class ChatGPT:
         answer = None
         while answer is None:
             try:
-                if self.model in {"gpt-3.5-turbo", "gpt-4"}:
+                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"}:
                     messages = self._generate_prompt_messages(message, dialog_messages, chat_mode)
                     r = await openai.ChatCompletion.acreate(
                         model=self.model,
@@ -67,7 +67,7 @@ class ChatGPT:
         answer = None
         while answer is None:
             try:
-                if self.model in {"gpt-3.5-turbo", "gpt-4"}:
+                if self.model in {"gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"}:
                     messages = self._generate_prompt_messages(message, dialog_messages, chat_mode)
                     r_gen = await openai.ChatCompletion.acreate(
                         model=self.model,
@@ -146,9 +146,12 @@ class ChatGPT:
     def _count_tokens_from_messages(self, messages, answer, model="gpt-3.5-turbo"):
         encoding = tiktoken.encoding_for_model(model)
 
-        if model == "gpt-3.5-turbo":
+        if model == "gpt-3.5-turbo-16k":
             tokens_per_message = 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n
             tokens_per_name = -1  # if there's a name, the role is omitted
+        elif model == "gpt-3.5-turbo":
+            tokens_per_message = 4
+            tokens_per_name = -1    
         elif model == "gpt-4":
             tokens_per_message = 3
             tokens_per_name = 1
diff --git a/config/models.yml b/config/models.yml
index 70f423b..4d522c5 100644
--- a/config/models.yml
+++ b/config/models.yml
@@ -1,4 +1,4 @@
-available_text_models: ["gpt-3.5-turbo", "gpt-4", "text-davinci-003"]
+available_text_models: ["gpt-3.5-turbo", "gpt-3.5-turbo-16k", "gpt-4", "text-davinci-003"]
 
 info:
   gpt-3.5-turbo:
@@ -6,7 +6,7 @@ info:
     name: ChatGPT
     description: ChatGPT is that well-known model. It's <b>fast</b> and <b>cheap</b>. Ideal for everyday tasks. If there are some tasks it can't handle, try the <b>GPT-4</b>.
 
-    price_per_1000_input_tokens: 0.002
+    price_per_1000_input_tokens: 0.0015
     price_per_1000_output_tokens: 0.002
 
     scores:
@@ -14,6 +14,19 @@ info:
       Fast: 5
       Cheap: 5
 
+  gpt-3.5-turbo-16k:
+    type: chat_completion
+    name: GPT-16K
+    description: ChatGPT is that well-known model. It's <b>fast</b> and <b>cheap</b>. Ideal for everyday tasks. If there are some tasks it can't handle, try the <b>GPT-4</b>.
+
+    price_per_1000_input_tokens: 0.003
+    price_per_1000_output_tokens: 0.004
+
+    scores:
+      Smart: 3
+      Fast: 5
+      Cheap: 5
+
   gpt-4:
     type: chat_completion
     name: GPT-4