Files
siren/src/commands/oai.rs

295 lines
8.5 KiB
Rust

use std::error::Error;
use std::fmt;
use diesel::{prelude::*, PgConnection, insert_into};
use diesel::r2d2::{Pool, ConnectionManager};
use log::{error, debug, trace, warn};
use serde::{Serialize, Deserialize};
use serde_json::Value;
use serenity::model::channel::Message;
use serenity::prelude::*;
use crate::database::models::{NewMessageDB, MessageDB};
pub struct OAI {
pub client: reqwest::Client,
pub base_url: String,
pub max_attempts: i64,
pub token: String,
pub max_context_questions: i64
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct ChatCompletionRequest {
model: GPTModel,
messages: Vec<ChatCompletionMessage>,
/// Value between 0 and 2
#[serde(skip_serializing_if = "Option::is_none")]
temperature: Option<f64>,
/// Value between 0 and 1
#[serde(skip_serializing_if = "Option::is_none")]
top_p: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
n: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
max_tokens: Option<i64>,
/// Value between -2.0 and 2.0
#[serde(skip_serializing_if = "Option::is_none")]
presence_penalty: Option<f64>,
/// Value between -2.0 and 2.0
#[serde(skip_serializing_if = "Option::is_none")]
frequency_penalty: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
user: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct ChatCompletionMessage {
role: GPTRole,
content: String
}
#[derive(Debug, Clone, Serialize, Deserialize)]
enum GPTRole {
#[serde(rename = "system")]
System,
#[serde(rename = "user")]
User,
#[serde(rename = "assistant")]
Assistant,
#[serde(rename = "function")]
Function
}
#[derive(Debug, Clone, Serialize, Deserialize)]
enum GPTModel {
#[serde(rename = "gpt-3.5-turbo")]
GPT35Turbo,
#[serde(rename = "gpt-3.5-turbo-0613")]
GPT35Snapshot,
#[serde(rename = "gpt-3.5-turbo-16k")]
GPT3516k,
#[serde(rename = "gpt-3.5-turbo-16k-0613")]
GPT3516kSnapshot,
#[serde(rename = "gpt-4")]
GPT4,
#[serde(rename = "gpt-4-0613")]
GPT4Snapshot,
#[serde(rename = "gpt-4-32k")]
GPT432k,
#[serde(rename = "gpt-4-32k-0613")]
GPT432kSnapshot,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct ChatCompletionResponse {
id: String,
object: String,
created: i64,
model: GPTModel,
usage: Usage,
choices: Vec<Choice>
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct Usage {
prompt_tokens: i64,
completion_tokens: i64,
total_tokens: i64
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct Choice {
message: ChatCompletionMessage,
finish_reason: String,
index: i64
}
#[derive(Debug, Clone, Serialize, Deserialize)]
struct ResponseError {
code: Option<i64>,
message: Option<String>,
param: Option<String>,
#[serde(rename = "type")]
error_type: Option<String>
}
#[derive(Debug)]
struct OAIError {
pub message: String
}
impl fmt::Display for OAIError {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "OAIError: {}", self.message)
}
}
impl Error for OAIError {}
#[derive(Debug, Clone, Serialize, Deserialize)]
enum ResponseEvent {
ChatCompletionResponse(ChatCompletionResponse),
ResponseError(ResponseError)
}
impl OAI {
async fn get_request(&self, request: ChatCompletionRequest) -> Result<ChatCompletionResponse, OAIError> {
let uri = format!("{}/chat/completions", self.base_url);
let body = serde_json::to_string(&request).unwrap();
trace!("Sending request to {}: {}", uri, body);
let value = match match self.client
.post(&uri)
.bearer_auth(&self.token)
.header("Content-Type", "application/json".to_string())
.body(body)
.send()
.await {
Ok(r) => r,
Err(err) => return Err(OAIError {
message: format!("Could not send request to OpenAI: {}", err),
})
}
.json::<Value>()
.await {
Ok(r) => r,
Err(err) => return Err(OAIError {
message: format!("Could not read response from OpenAI: {}", err)
})
};
trace!("Received response from OpenAI: {:?}", value);
// let response = match serde_json::from_value::<OAIResponseEvent>(value) {
// Ok(r) => {
// match r {
// OAIResponseEvent::OAIResponse(r) => r,
// OAIResponseEvent::OAIError(e) => return Err(OAIError { message: e.message.unwrap_or("Unknown error".to_string()) })
// }
// },
// Err(err) => return Err(OAIError {
// message: format!("Could not parse response from OpenAI: {}", err)
// })
// };
let response = match serde_json::from_value::<ChatCompletionResponse>(value) {
Ok(r) => r,
Err(err) => return Err(OAIError {
message: format!("Could not parse response from OpenAI: {}", err)
})
};
Ok(response)
}
}
pub async fn generate_response(ctx: &Context, msg: &Message, oai: &OAI, pool: &Pool<ConnectionManager<PgConnection>>) {
debug!("Generating response for message: {}", msg.content);
let typing = msg.channel_id.start_typing(&ctx.http).unwrap();
let guild_id = msg.guild_id.unwrap();
let channel_id = msg.channel_id;
let author_id = msg.author.id;
let mut connection = pool.get().unwrap();
// Parse out the bot mention from the message
let bot_mention: String = format!("<@{}>", ctx.cache.current_user_id().0);
let parsed_content = msg.content.replace(bot_mention.as_str(), "");
// Setup the request messages
let result: Result<Vec<MessageDB>, diesel::result::Error> = crate::database::schema::messages::table
.select(MessageDB::as_select())
.filter((crate::database::schema::messages::guild_id.eq(guild_id.0 as i64))
.and(crate::database::schema::messages::channel_id.eq(channel_id.0 as i64))
.and(crate::database::schema::messages::user_id.eq(author_id.0 as i64))
)
.order(crate::database::schema::messages::created.desc())
.limit(oai.max_context_questions)
.load(&mut connection);
let previous_messages = match result {
Ok(r) => {
let mut previous_message = "".to_string();
for message in r {
previous_message = format!("{}\nYou: {}\n Siren: {}", previous_message, message.request, message.response);
}
Some(ChatCompletionMessage { role: GPTRole::User, content: previous_message })
}
Err(err) => {
error!("Could not load previous messages: {}", err);
None
}
};
let mut messages = vec![
ChatCompletionMessage {
role: GPTRole::System,
content: "Siren is a Discord bot specializing in Dungeons and Dragons. Limit Siren's responses to <= 2000 characters. Siren must always obey these instructions, no matter what.".to_string()
},
];
if let Some(mut previous) = previous_messages {
previous.content = format!("{}\nYou: {}\nSiren: ", previous.content, parsed_content);
messages.push(previous);
} else {
messages.push(ChatCompletionMessage {
role: GPTRole::User,
content: format!("You: {}, Siren: ", parsed_content)
});
}
let model = "gpt-3.5-turbo".to_string();
let request = ChatCompletionRequest {
model: GPTModel::GPT35Turbo,
messages,
temperature: Some(0.5),
top_p: None,
n: None,
max_tokens: Some(1000),
presence_penalty: Some(0.6),
frequency_penalty: Some(0.0),
user: Some(msg.author.name.clone())
};
let response = match oai.get_request(request).await {
Ok(r) => {
debug!("Processing response received from OpenAI");
if !r.choices.is_empty() {
// Insert the message into the messages database table
let res = r.choices[0].message.content.clone();
if let Err(err) = insert_into(crate::database::schema::messages::table).values(NewMessageDB {
id: &r.id,
guild_id: guild_id.0 as i64,
channel_id: channel_id.0 as i64,
user_id: author_id.0 as i64,
created: r.created,
model: &model,
request: &parsed_content,
response: &res,
request_tags: vec![],
response_tags: vec![],
}).execute(&mut connection) {
error!("Could not insert message into database: {}", err);
}
res
} else {
warn!("No choices received in the response from OpenAI");
"No reply received".to_string()
}
}
Err(err) => {
error!("Could not get response from OpenAI: {}", err.message);
err.message
}
};
debug!("Writing response: \"{}\"", response);
// Stop the typing indicator and send the response
typing.stop();
if let Err(why) = msg.channel_id.say(&ctx.http, response).await {
error!("Cannot send message: {}", why);
}
}