Merge branch 'main' into gpt-index

Signed-off-by: Kaveen Kumarasinghe <k5kumara@uwaterloo.ca>

cogs/commands.py

@@ -309,6 +309,9 @@ class Commands(discord.Cog, name="Commands"):
     @discord.option(
         name="prompt", description="The prompt to send to GPT3", required=True
     )
+    @discord.option(
+        name="private", description="Will only be visible to you", required=False
+    )
     @discord.option(
         name="temperature",
         description="Higher values means the model will take more risks",
@@ -342,13 +345,20 @@ class Commands(discord.Cog, name="Commands"):
         self,
         ctx: discord.ApplicationContext,
         prompt: str,
+        private: bool,
         temperature: float,
         top_p: float,
         frequency_penalty: float,
         presence_penalty: float,
     ):
         await self.converser_cog.ask_command(
-            ctx, prompt, temperature, top_p, frequency_penalty, presence_penalty
+            ctx,
+            prompt,
+            private,
+            temperature,
+            top_p,
+            frequency_penalty,
+            presence_penalty,
         )
 
     @add_to_group("gpt")
@@ -368,6 +378,9 @@ class Commands(discord.Cog, name="Commands"):
         required=False,
         default="",
     )
+    @discord.option(
+        name="private", description="Will only be visible to you", required=False
+    )
     @discord.option(
         name="temperature",
         description="Higher values means the model will take more risks",
@@ -393,12 +406,13 @@ class Commands(discord.Cog, name="Commands"):
         ctx: discord.ApplicationContext,
         instruction: str,
         text: str,
+        private: bool,
         temperature: float,
         top_p: float,
         codex: bool,
     ):
         await self.converser_cog.edit_command(
-            ctx, instruction, text, temperature, top_p, codex
+            ctx, instruction, text, private, temperature, top_p, codex
         )
 
     @add_to_group("gpt")
@@ -430,6 +444,13 @@ class Commands(discord.Cog, name="Commands"):
         required=False,
         default=False,
     )
+    @discord.option(
+        name="model",
+        description="Which model to use with the bot",
+        required=False,
+        default=False,
+        autocomplete=Settings_autocompleter.get_models,
+    )
     @discord.option(
         name="temperature",
         description="Higher values means the model will take more risks",
@@ -470,6 +491,7 @@ class Commands(discord.Cog, name="Commands"):
         opener_file: str,
         private: bool,
         minimal: bool,
+        model: str,
         temperature: float,
         top_p: float,
         frequency_penalty: float,
@@ -481,6 +503,7 @@ class Commands(discord.Cog, name="Commands"):
             opener_file,
             private,
             minimal,
+            model,
             temperature,
             top_p,
             frequency_penalty,

cogs/text_service_cog.py

@@ -693,6 +693,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
         self,
         ctx: discord.ApplicationContext,
         prompt: str,
+        private: bool,
         temperature: float,
         top_p: float,
         frequency_penalty: float,
@@ -720,7 +721,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
             if not user_api_key:
                 return
 
-        await ctx.defer()
+        await ctx.defer(ephemeral=private)
 
         overrides = Override(temperature, top_p, frequency_penalty, presence_penalty)
 
@@ -741,6 +742,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
         ctx: discord.ApplicationContext,
         instruction: str,
         text: str,
+        private: bool,
         temperature: float,
         top_p: float,
         codex: bool,
@@ -766,7 +768,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
             if not user_api_key:
                 return
 
-        await ctx.defer()
+        await ctx.defer(ephemeral=private)
 
         overrides = Override(temperature, top_p, 0, 0)
 
@@ -801,6 +803,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
         opener_file: str,
         private: bool,
         minimal: bool,
+        model: str,
         temperature: float,
         top_p: float,
         frequency_penalty: float,
@@ -814,6 +817,7 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
             opener_file (str): A .txt or .json file which is appended before the opener
             private (bool): If the thread should be private
             minimal (bool): If a minimal starter should be used
+            model (str): The openai model that should be used
             temperature (float): Sets the temperature override
             top_p (float): Sets the top p override
             frequency_penalty (float): Sets the frequency penalty override
@@ -866,7 +870,9 @@ class GPT3ComCon(discord.Cog, name="GPT3ComCon"):
             )
 
         self.conversation_threads[thread.id] = Thread(thread.id)
-        self.conversation_threads[thread.id].model = self.model.model
+        self.conversation_threads[thread.id].model = (
+            self.model.model if not model else model
+        )
 
         # Set the overrides for the conversation
         self.conversation_threads[thread.id].set_overrides(

detailed_guides/azure-app-service.md

@@ -0,0 +1,58 @@
+<img src="https://user-images.githubusercontent.com/39274208/216566367-e1311c9e-bbc1-42fe-9e09-69e9fb728133.png" width="300">
+
+You can choose to host the docker contained on app service so you have less maintenance, and it is also cheaper than a VM. 
+As a student, you can get free Azure credits and you can get free Azure credits with you MSDN license if you work at a Microsoft partner.
+The steps are also fairly straightforward.
+
+To make it work on Azure App service follow the following steps: 
+
+<img src="https://user-images.githubusercontent.com/39274208/216566481-e06ccf0d-7346-438b-ab5a-7cd4ac6b08f5.png" width="150">
+
+Creating the container: 
+
+1. Sign into Azure via portal.azure.com
+2. Search "app services" and click on it
+3. choose resource group, webapp name.
+4. At "publish" choose: "docker container"
+5. OS: linux
+6. Region: whichever is closer to you
+7. choose Basic B1 as app service plan. this will cost you around 10 dollar in credits
+8. At the "docker tab "change image source" to "dockerhub"
+9. At image, use the following: kaveenk/gpt3discord:latest_release
+10. Go to "review and create" and create the container! 
+11. Wait a minute until it gets provisioned. 
+
+<img src="https://user-images.githubusercontent.com/39274208/216566567-a00bd5c1-1ab4-4250-a6d6-9ab9d872bd31.png" width="200">
+
+Configurating the container:
+
+App services makes it easy for you to add ENV files without using the CLI. We do this first
+
+1. Go to "configuration" under "settings"
+Enter the following env files as defined in the readme.md file in this repository: 
+
+The most important application settings are:
+ADMIN_ROLES
+ALLOWED_GUILDS (your discord server token! not to be confused with a channel token or bot token)
+DALLE_ROLES
+DEBUG_CHANNEL
+DEBUG_GUILD
+DISCORD_TOKEN
+GPT_ROLES
+OPENAI_TOKEN
+HEALTH_SERVICE_ENABLED
+
+<img src="https://user-images.githubusercontent.com/39274208/216571201-1f5f96e8-b2f9-4e45-8b29-89c2bc3be97e.png" width="200">
+Health:
+
+ 1. It's important to also add WEBSITES_PORT to the application settings, since we are using the port for the health check of the container. Use 8181.
+ 2. Go to "Health check" under monitoring and enter the following value: /healthz
+ 3. Of course, set health check to: "on"
+ 4. Press save.
+
+Last step: go to "deployment center" and turn off "continious deployment". 
+This will make sure we get the latest build from docker hub when we restart the container. 
+
+It's as easy as this. Of course, make sure to follow the other steps settings up the discord bot in the readme.md.
+
+Hope this helps! 

models/autocomplete_model.py

@@ -86,6 +86,14 @@ class Settings_autocompleter:
         await ctx.interaction.response.defer()  # defer so the autocomplete in int values doesn't error but rather just says not found
         return []
 
+    async def get_models(
+        ctx: discord.AutocompleteContext,
+    ):
+        """Gets all models"""
+        return [
+            value for value in Models.TEXT_MODELS if value.startswith(ctx.value.lower())
+        ]
+
     async def get_value_moderations(
         ctx: discord.AutocompleteContext,
     ):  # Behaves a bit weird if you go back and edit the parameter without typing in a new command

models/openai_model.py

@@ -78,7 +78,7 @@ class Models:
 
     @staticmethod
     def get_max_tokens(model: str) -> int:
-        return Models.TOKEN_MAPPING.get(model, 4024)
+        return Models.TOKEN_MAPPING.get(model, 2024)
 
 
 class ImageSize:
@@ -576,12 +576,18 @@ class Model:
         self._prompt_min_length = value
         SETTINGS_DB["prompt_min_length"] = value
 
-    def backoff_handler(details):
+    def backoff_handler_http(details):
         print(
             f"Backing off {details['wait']:0.1f} seconds after {details['tries']} tries calling function {details['target']} | "
             f"{details['exception'].status}: {details['exception'].message}"
         )
 
+    def backoff_handler_request(details):
+        print(
+            f"Backing off {details['wait']:0.1f} seconds after {details['tries']} tries calling function {details['target']} | "
+            f"{details['exception'].args[0]}"
+        )
+
     async def valid_text_request(self, response):
         try:
             tokens_used = int(response["usage"]["total_tokens"])
@@ -598,7 +604,7 @@ class Model:
         factor=3,
         base=5,
         max_tries=4,
-        on_backoff=backoff_handler,
+        on_backoff=backoff_handler_http,
     )
     async def send_embedding_request(self, text, custom_api_key=None):
         async with aiohttp.ClientSession(raise_for_status=True) as session:
@@ -624,11 +630,11 @@ class Model:
 
     @backoff.on_exception(
         backoff.expo,
-        aiohttp.ClientResponseError,
+        ValueError,
         factor=3,
         base=5,
-        max_tries=6,
-        on_backoff=backoff_handler,
+        max_tries=4,
+        on_backoff=backoff_handler_request,
     )
     async def send_edit_request(
         self,
@@ -651,7 +657,7 @@ class Model:
         )
         print(f"Overrides -> temp:{temp_override}, top_p:{top_p_override}")
 
-        async with aiohttp.ClientSession(raise_for_status=True) as session:
+        async with aiohttp.ClientSession(raise_for_status=False) as session:
             payload = {
                 "model": Models.EDIT if codex is False else Models.CODE_EDIT,
                 "input": "" if text is None else text,
@@ -676,7 +682,7 @@ class Model:
         factor=3,
         base=5,
         max_tries=6,
-        on_backoff=backoff_handler,
+        on_backoff=backoff_handler_http,
     )
     async def send_moderations_request(self, text):
         # Use aiohttp to send the above request:
@@ -695,11 +701,11 @@ class Model:
 
     @backoff.on_exception(
         backoff.expo,
-        aiohttp.ClientResponseError,
+        ValueError,
         factor=3,
         base=5,
         max_tries=4,
-        on_backoff=backoff_handler,
+        on_backoff=backoff_handler_request,
     )
     async def send_summary_request(self, prompt, custom_api_key=None):
         """
@@ -718,7 +724,7 @@ class Model:
 
         tokens = self.usage_service.count_tokens(summary_request_text)
 
-        async with aiohttp.ClientSession(raise_for_status=True) as session:
+        async with aiohttp.ClientSession(raise_for_status=False) as session:
             payload = {
                 "model": Models.DAVINCI,
                 "prompt": summary_request_text,
@@ -746,11 +752,11 @@ class Model:
 
     @backoff.on_exception(
         backoff.expo,
-        aiohttp.ClientResponseError,
+        ValueError,
         factor=3,
         base=5,
         max_tries=4,
-        on_backoff=backoff_handler,
+        on_backoff=backoff_handler_request,
     )
     async def send_request(
         self,
@@ -774,12 +780,16 @@ class Model:
                 f"Prompt must be greater than {self.prompt_min_length} characters, it is currently: {len(prompt)} characters"
             )
 
+        if not max_tokens_override:
+            if model:
+                max_tokens_override = Models.get_max_tokens(model) - tokens
+
         print(f"The prompt about to be sent is {prompt}")
         print(
             f"Overrides -> temp:{temp_override}, top_p:{top_p_override} frequency:{frequency_penalty_override}, presence:{presence_penalty_override}"
         )
 
-        async with aiohttp.ClientSession(raise_for_status=True) as session:
+        async with aiohttp.ClientSession(raise_for_status=False) as session:
             payload = {
                 "model": self.model if model is None else model,
                 "prompt": prompt,
@@ -787,7 +797,7 @@ class Model:
                 "temperature": self.temp if temp_override is None else temp_override,
                 "top_p": self.top_p if top_p_override is None else top_p_override,
                 "max_tokens": self.max_tokens - tokens
-                if not max_tokens_override
+                if max_tokens_override is None
                 else max_tokens_override,
                 "presence_penalty": self.presence_penalty
                 if presence_penalty_override is None
@@ -839,7 +849,7 @@ class Model:
         factor=3,
         base=5,
         max_tries=4,
-        on_backoff=backoff_handler,
+        on_backoff=backoff_handler_http,
     )
     async def send_image_request(
         self, ctx, prompt, vary=None, custom_api_key=None