feat: add an ability to set default model through an env config (#70)

* feat: use zod enums for models list

* feat: use env to set default LLM models

* feat: make models input optional

.env.example

@@ -1,5 +1,6 @@
-# AI SERVER
+# AI SERVER CONFIGS
 AI_SERVER_URL="http://localhost:4000"
+DEFAULT_MODEL="tinydolphin"
 
 # BOT CONFIGS
 CLIENT_ID=CLIENT_ID_GOES_HERE

.gitignore

@@ -11,4 +11,4 @@ dist
 
 # Env
 .env*
-!.env.sample
+!.env.example

src/autocompletes/select-model/autocomplete.ts

@@ -1,12 +1,20 @@
 import { SUPPORTED_MODELS_MAP } from '../../llm/utils';
 import type { AutocompleteHandler } from '../builder';
 
+function get25ptions(models: typeof SUPPORTED_MODELS_MAP) {
+  return models.slice(0, 25);
+}
+
 export const selectModelAutocomplete: AutocompleteHandler = async (interaction) => {
-  const searchTerm = interaction.options.getString('model', true).trim().toLowerCase();
+  let searchTerm = interaction.options.getString('model', false);
+  if (!searchTerm) {
+    const options = get25ptions(SUPPORTED_MODELS_MAP);
+    interaction.respond(options);
+    return;
+  }
 
-  const options = SUPPORTED_MODELS_MAP.filter((model) => {
-    if (!searchTerm) return true;
-    return model.name.includes(searchTerm);
-  }).slice(0, 25);
+  searchTerm = searchTerm.trim().toLowerCase();
+  const filtered = SUPPORTED_MODELS_MAP.filter((model) => model.name.includes(searchTerm));
+  const options = get25ptions(filtered);
   interaction.respond(options);
 };

src/llm/utils.ts

@@ -1,14 +1,15 @@
 import { Result } from 'oxide.ts';
+import { z } from 'zod';
 import { logger } from '../utils/logger';
 import { getClient } from './client';
 
 const DISCORD_MESSAGE_MAX_CHARACTERS = 2000;
 const QUESTION_CUT_OFF_LENGTH = 150;
 const RESERVED_LENGTH = 50; // for other additional strings. E.g. number `(1/4)`, `Q: `, `A: `, etc.
 
-const SUPPORTED_MODELS = ['gpt-3.5-turbo', 'gpt-4', 'phi', 'phi3', 'tinydolphin', 'mistral', 'mixtral', 'llama3', 'llama3-70b'] as const;
-type SupportedModel = (typeof SUPPORTED_MODELS)[number];
-export const SUPPORTED_MODELS_MAP = SUPPORTED_MODELS.map((model) => ({
+export const SupportedModel = z.enum(['gpt-3.5-turbo', 'gpt-4', 'phi', 'phi3', 'tinydolphin', 'mistral', 'mixtral', 'llama3', 'llama3-70b']);
+export type SupportedModel = z.infer<typeof SupportedModel>;
+export const SUPPORTED_MODELS_MAP = SupportedModel.options.map((model) => ({
   name: model,
   value: model,
 }));
@@ -84,10 +85,10 @@ function addNumber(chunks: string[]): string[] {
   return output;
 }
 
-export function findModel(model: string): SupportedModel {
-  const hasModel = SUPPORTED_MODELS.find((option) => option.toLowerCase() === model);
-  if (!hasModel) {
-    throw new Error(`Model ${model} is not supported. Supported models are: ${SUPPORTED_MODELS.join(', ')}`);
+export function findModel(input: string): SupportedModel {
+  const model = SupportedModel.safeParse(input);
+  if (!model.success) {
+    throw new Error(`Model ${input} is not supported. Supported models are: ${SupportedModel.options.join(', ')}`);
   }
-  return model as SupportedModel;
+  return model.data;
 }

src/slash-commands/ask-llm/command.ts

@@ -8,12 +8,12 @@ import type { SlashCommand, SlashCommandHandler } from '../builder';
 const data = new SlashCommandBuilder()
   .setName('ask')
   .setDescription('Ask an LLM to answer anything')
-  .addStringOption((option) => option.setName('model').setDescription('Choose an LLM model').setRequired(true).setAutocomplete(true))
+  .addStringOption((option) => option.setName('model').setDescription('Choose an LLM model').setRequired(false).setAutocomplete(true))
   .addStringOption((option) => option.setName('question').setDescription('Enter your prompt').setRequired(true).setMinLength(10));
 
 export const execute: SlashCommandHandler = async (interaction) => {
   await interaction.deferReply();
-  const model = interaction.options.getString('model', true).trim().toLowerCase();
+  const model = (interaction.options.getString('model', false) || process.env.DEFAULT_MODEL).trim().toLowerCase();
   const question = interaction.options.getString('question', true).trim();
   logger.info(`[ask]: Asking ${model} model with prompt: ${question}`);
 

src/slash-commands/review-resume/command.ts

@@ -11,12 +11,12 @@ import { getOutputFileName } from './utils';
 const data = new SlashCommandBuilder()
   .setName('review-resume')
   .setDescription('Review a resume from a generic PDF URL or Google Drive')
-  .addStringOption((option) => option.setName('model').setDescription('Choose an LLM model').setRequired(true).setAutocomplete(true))
+  .addStringOption((option) => option.setName('model').setDescription('Choose an LLM model').setRequired(false).setAutocomplete(true))
   .addStringOption((option) => option.setName('url').setDescription('PDF or Google Drive URL').setRequired(true));
 
 export const execute: SlashCommandHandler = async (interaction) => {
   await interaction.deferReply();
-  const model = interaction.options.getString('model', true).trim().toLowerCase();
+  const model = (interaction.options.getString('model', false) || process.env.DEFAULT_MODEL).trim().toLowerCase();
   const url = interaction.options.getString('url', true);
   logger.info(`[review-resume]: Reviewing resume from URL: ${url}`);
 

src/utils/load-env.ts

@@ -1,5 +1,6 @@
 import dotenv from 'dotenv';
 import { z } from 'zod';
+import { SupportedModel } from '../llm/utils';
 import { logger } from './logger';
 
 const configSchema = z.object({
@@ -10,8 +11,9 @@ const configSchema = z.object({
   CLIENT_ID: z.string(),
   GUILD_ID: z.string().optional(),
 
-  // AI Server URL
+  // AI Server config
   AI_SERVER_URL: z.string().url(),
+  DEFAULT_MODEL: SupportedModel,
 });
 type ConfigSchema = z.infer<typeof configSchema>;