FEATURE: Configurable LLMs. (#606)

This PR introduces the concept of "LlmModel" as a new way to quickly add new LLM models without making any code changes. We are releasing this first version and will add incremental improvements, so expect changes.

The AI Bot can't fully take advantage of this feature as users are hard-coded. We'll fix this in a separate PR.s

admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-new.js

@@ -0,0 +1,16 @@
+import DiscourseRoute from "discourse/routes/discourse";
+
+export default DiscourseRoute.extend({
+  async model() {
+    const record = this.store.createRecord("ai-llm");
+    return record;
+  },
+
+  setupController(controller, model) {
+    this._super(controller, model);
+    controller.set(
+      "allLlms",
+      this.modelFor("adminPlugins.show.discourse-ai-llms")
+    );
+  },
+});

admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-show.js

@@ -0,0 +1,17 @@
+import DiscourseRoute from "discourse/routes/discourse";
+
+export default DiscourseRoute.extend({
+  async model(params) {
+    const allLlms = this.modelFor("adminPlugins.show.discourse-ai-llms");
+    const id = parseInt(params.id, 10);
+    return allLlms.findBy("id", id);
+  },
+
+  setupController(controller, model) {
+    this._super(controller, model);
+    controller.set(
+      "allLlms",
+      this.modelFor("adminPlugins.show.discourse-ai-llms")
+    );
+  },
+});

admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms.js

@@ -0,0 +1,7 @@
+import DiscourseRoute from "discourse/routes/discourse";
+
+export default class DiscourseAiAiLlmsRoute extends DiscourseRoute {
+  model() {
+    return this.store.findAll("ai-llm");
+  }
+}

admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-llms/index.hbs

@@ -0,0 +1 @@
+<AiLlmsListEditor @llms={{this.model}} />

admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-llms/new.hbs

@@ -0,0 +1 @@
+<AiLlmsListEditor @llms={{this.allLlms}} @currentLlm={{this.model}} />

admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-llms/show.hbs

@@ -0,0 +1 @@
+<AiLlmsListEditor @llms={{this.allLlms}} @currentLlm={{this.model}} />

app/controllers/discourse_ai/admin/ai_llms_controller.rb

@@ -0,0 +1,64 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module Admin
+    class AiLlmsController < ::Admin::AdminController
+      requires_plugin ::DiscourseAi::PLUGIN_NAME
+
+      def index
+        llms = LlmModel.all
+
+        render json: {
+                 ai_llms:
+                   ActiveModel::ArraySerializer.new(
+                     llms,
+                     each_serializer: LlmModelSerializer,
+                     root: false,
+                   ).as_json,
+                 meta: {
+                   providers: DiscourseAi::Completions::Llm.provider_names,
+                   tokenizers:
+                     DiscourseAi::Completions::Llm.tokenizer_names.map { |tn|
+                       { id: tn, name: tn.split("::").last }
+                     },
+                 },
+               }
+      end
+
+      def show
+        llm_model = LlmModel.find(params[:id])
+        render json: LlmModelSerializer.new(llm_model)
+      end
+
+      def create
+        if llm_model = LlmModel.new(ai_llm_params).save
+          render json: { ai_persona: llm_model }, status: :created
+        else
+          render_json_error llm_model
+        end
+      end
+
+      def update
+        llm_model = LlmModel.find(params[:id])
+
+        if llm_model.update(ai_llm_params)
+          render json: llm_model
+        else
+          render_json_error llm_model
+        end
+      end
+
+      private
+
+      def ai_llm_params
+        params.require(:ai_llm).permit(
+          :display_name,
+          :name,
+          :provider,
+          :tokenizer,
+          :max_prompt_tokens,
+        )
+      end
+    end
+  end
+end

app/models/llm_model.rb

@@ -0,0 +1,7 @@
+# frozen_string_literal: true
+
+class LlmModel < ActiveRecord::Base
+  def tokenizer_class
+    tokenizer.constantize
+  end
+end

app/serializers/llm_model_serializer.rb

@@ -0,0 +1,7 @@
+# frozen_string_literal: true
+
+class LlmModelSerializer < ApplicationSerializer
+  root "llm"
+
+  attributes :id, :display_name, :name, :provider, :max_prompt_tokens, :tokenizer
+end

assets/javascripts/discourse/admin-discourse-ai-plugin-route-map.js

@@ -8,5 +8,10 @@ export default {
       this.route("new");
       this.route("show", { path: "/:id" });
     });
+
+    this.route("discourse-ai-llms", { path: "ai-llms" }, function () {
+      this.route("new");
+      this.route("show", { path: "/:id" });
+    });
   },
 };

assets/javascripts/discourse/admin/adapters/ai-llm.js

@@ -0,0 +1,21 @@
+import RestAdapter from "discourse/adapters/rest";
+
+export default class Adapter extends RestAdapter {
+  jsonMode = true;
+
+  basePath() {
+    return "/admin/plugins/discourse-ai/";
+  }
+
+  pathFor(store, type, findArgs) {
+    // removes underscores which are implemented in base
+    let path =
+      this.basePath(store, type, findArgs) +
+      store.pluralize(this.apiNameFor(type));
+    return this.appendQueryParams(path, findArgs);
+  }
+
+  apiNameFor() {
+    return "ai-llm";
+  }
+}

assets/javascripts/discourse/admin/models/ai-llm.js

@@ -0,0 +1,20 @@
+import RestModel from "discourse/models/rest";
+
+export default class AiLlm extends RestModel {
+  createProperties() {
+    return this.getProperties(
+      "display_name",
+      "name",
+      "provider",
+      "tokenizer",
+      "max_prompt_tokens"
+    );
+  }
+
+  updateProperties() {
+    const attrs = this.createProperties();
+    attrs.id = this.id;
+
+    return attrs;
+  }
+}

assets/javascripts/discourse/components/ai-llm-editor.gjs

@@ -0,0 +1,122 @@
+import Component from "@glimmer/component";
+import { tracked } from "@glimmer/tracking";
+import { Input } from "@ember/component";
+import { action } from "@ember/object";
+import { later } from "@ember/runloop";
+import { inject as service } from "@ember/service";
+import DButton from "discourse/components/d-button";
+import { popupAjaxError } from "discourse/lib/ajax-error";
+import i18n from "discourse-common/helpers/i18n";
+import I18n from "discourse-i18n";
+import ComboBox from "select-kit/components/combo-box";
+import DTooltip from "float-kit/components/d-tooltip";
+
+export default class AiLlmEditor extends Component {
+  @service toasts;
+  @service router;
+
+  @tracked isSaving = false;
+
+  get selectedProviders() {
+    const t = (provName) => {
+      return I18n.t(`discourse_ai.llms.providers.${provName}`);
+    };
+
+    return this.args.llms.resultSetMeta.providers.map((prov) => {
+      return { id: prov, name: t(prov) };
+    });
+  }
+
+  @action
+  async save() {
+    this.isSaving = true;
+    const isNew = this.args.model.isNew;
+
+    try {
+      await this.args.model.save();
+
+      if (isNew) {
+        this.args.llms.addObject(this.args.model);
+        this.router.transitionTo(
+          "adminPlugins.show.discourse-ai-llms.show",
+          this.args.model
+        );
+      } else {
+        this.toasts.success({
+          data: { message: I18n.t("discourse_ai.llms.saved") },
+          duration: 2000,
+        });
+      }
+    } catch (e) {
+      popupAjaxError(e);
+    } finally {
+      later(() => {
+        this.isSaving = false;
+      }, 1000);
+    }
+  }
+
+  <template>
+    <form class="form-horizontal ai-llm-editor">
+      <div class="control-group">
+        <label>{{i18n "discourse_ai.llms.display_name"}}</label>
+        <Input
+          class="ai-llm-editor__display-name"
+          @type="text"
+          @value={{@model.display_name}}
+        />
+      </div>
+      <div class="control-group">
+        <label>{{i18n "discourse_ai.llms.name"}}</label>
+        <Input
+          class="ai-llm-editor__name"
+          @type="text"
+          @value={{@model.name}}
+        />
+        <DTooltip
+          @icon="question-circle"
+          @content={{I18n.t "discourse_ai.llms.hints.name"}}
+        />
+      </div>
+      <div class="control-group">
+        <label>{{I18n.t "discourse_ai.llms.provider"}}</label>
+        <ComboBox
+          @value={{@model.provider}}
+          @content={{this.selectedProviders}}
+        />
+      </div>
+      <div class="control-group">
+        <label>{{I18n.t "discourse_ai.llms.tokenizer"}}</label>
+        <ComboBox
+          @value={{@model.tokenizer}}
+          @content={{@llms.resultSetMeta.tokenizers}}
+        />
+      </div>
+      <div class="control-group">
+        <label>{{i18n "discourse_ai.llms.max_prompt_tokens"}}</label>
+        <Input
+          @type="number"
+          class="ai-llm-editor__max-prompt-tokens"
+          step="any"
+          min="0"
+          lang="en"
+          @value={{@model.max_prompt_tokens}}
+        />
+        <DTooltip
+          @icon="question-circle"
+          @content={{I18n.t "discourse_ai.llms.hints.max_prompt_tokens"}}
+        />
+      </div>
+
+      <div class="control-group ai-llm-editor__action_panel">
+        <DButton
+          class="btn-primary ai-llm-editor__save"
+          @action={{this.save}}
+          @disabled={{this.isSaving}}
+        >
+          {{I18n.t "discourse_ai.llms.save"}}
+        </DButton>
+      </div>
+    </form>
+  </template>
+}

assets/javascripts/discourse/components/ai-llms-list-editor.gjs

@@ -0,0 +1,61 @@
+import Component from "@glimmer/component";
+import { LinkTo } from "@ember/routing";
+import icon from "discourse-common/helpers/d-icon";
+import i18n from "discourse-common/helpers/i18n";
+import I18n from "discourse-i18n";
+import AiLlmEditor from "./ai-llm-editor";
+
+export default class AiLlmsListEditor extends Component {
+  get hasNoLLMElements() {
+    this.args.llms.length !== 0;
+  }
+
+  <template>
+    <section class="ai-llms-list-editor admin-detail pull-left">
+
+      <div class="ai-llms-list-editor__header">
+        <h3>{{i18n "discourse_ai.llms.short_title"}}</h3>
+        {{#unless @currentLlm.isNew}}
+          <LinkTo
+            @route="adminPlugins.show.discourse-ai-llms.new"
+            class="btn btn-small btn-primary"
+          >
+            {{icon "plus"}}
+            <span>{{I18n.t "discourse_ai.llms.new"}}</span>
+          </LinkTo>
+        {{/unless}}
+      </div>
+
+      <div class="ai-llms-list-editor__container">
+        {{#if this.hasNoLLMElements}}
+          <div class="ai-llms-list-editor__empty_list">
+            {{icon "robot"}}
+            {{i18n "discourse_ai.llms.no_llms"}}
+          </div>
+        {{else}}
+          <div class="content-list ai-llms-list-editor__content_list">
+            <ul>
+              {{#each @llms as |llm|}}
+                <li>
+                  <LinkTo
+                    @route="adminPlugins.show.discourse-ai-llms.show"
+                    current-when="true"
+                    @model={{llm}}
+                  >
+                    {{llm.display_name}}
+                  </LinkTo>
+                </li>
+              {{/each}}
+            </ul>
+          </div>
+        {{/if}}
+
+        <div class="ai-llms-list-editor__current">
+          {{#if @currentLlm}}
+            <AiLlmEditor @model={{@currentLlm}} @llms={{@llms}} />
+          {{/if}}
+        </div>
+      </div>
+    </section>
+  </template>
+}

assets/javascripts/initializers/admin-plugin-configuration-nav.js

@@ -16,6 +16,10 @@ export default {
           label: "discourse_ai.ai_persona.short_title",
           route: "adminPlugins.show.discourse-ai-personas",
         },
+        {
+          label: "discourse_ai.llms.short_title",
+          route: "adminPlugins.show.discourse-ai-llms",
+        },
       ]);
     });
   },

assets/stylesheets/modules/llms/common/ai-llms-editor.scss

@@ -0,0 +1,32 @@
+.ai-llms-list-editor {
+  &__header {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    margin: 0 0 1em 0;
+
+    h3 {
+      margin: 0;
+    }
+  }
+
+  &__container {
+    display: flex;
+    flex-direction: row;
+    align-items: center;
+    gap: 20px;
+    width: 100%;
+    align-items: stretch;
+  }
+
+  &__empty_list,
+  &__content_list {
+    min-width: 300px;
+  }
+
+  &__empty_list {
+    align-content: center;
+    text-align: center;
+    font-size: var(--font-up-1);
+  }
+}

config/locales/client.en.yml

@@ -194,6 +194,32 @@ en:
           uploading: "Uploading..."
           remove: "Remove upload"
 
+      llms:
+        short_title: "LLMs"
+        no_llms: "No LLMs yet"
+        new: "New Model"
+        display_name: "Name to display:"
+        name: "Model name:"
+        provider: "Service hosting the model:"
+        tokenizer: "Tokenizer:"
+        max_prompt_tokens: "Number of tokens for the prompt:"
+        save: "Save"
+        saved: "LLM Model Saved"
+
+        hints:
+          max_prompt_tokens: "Max numbers of tokens for the prompt. As a rule of thumb, this should be 50% of the model's context window."
+          name: "We include this in the API call to specify which model we'll use."
+
+        providers:
+          aws_bedrock: "AWS Bedrock"
+          anthropic: "Anthropic"
+          vllm: "vLLM"
+          hugging_face: "Hugging Face"
+          cohere: "Cohere"
+          open_ai: "OpenAI"
+          google: "Google"
+          azure: "Azure"
+
       related_topics:
         title: "Related Topics"
         pill: "Related"

config/routes.rb

@@ -45,5 +45,10 @@ Discourse::Application.routes.draw do
     post "/ai-personas/files/upload", to: "discourse_ai/admin/ai_personas#upload_file"
     put "/ai-personas/:id/files/remove", to: "discourse_ai/admin/ai_personas#remove_file"
     get "/ai-personas/:id/files/status", to: "discourse_ai/admin/ai_personas#indexing_status_check"
+
+    resources :ai_llms,
+              only: %i[index create show update],
+              path: "ai-llms",
+              controller: "discourse_ai/admin/ai_llms"
   end
 end

db/migrate/20240504222307_create_llm_model_table.rb

@@ -0,0 +1,14 @@
+# frozen_string_literal: true
+
+class CreateLlmModelTable < ActiveRecord::Migration[7.0]
+  def change
+    create_table :llm_models do |t|
+      t.string :display_name
+      t.string :name, null: false
+      t.string :provider, null: false
+      t.string :tokenizer, null: false
+      t.integer :max_prompt_tokens, null: false
+      t.timestamps
+    end
+  end
+end

lib/ai_bot/bot.rb

@@ -150,59 +150,73 @@ module DiscourseAi
 
       def self.guess_model(bot_user)
         # HACK(roman): We'll do this until we define how we represent different providers in the bot settings
-        case bot_user.id
-        when DiscourseAi::AiBot::EntryPoint::CLAUDE_V2_ID
-          if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-2")
-            "aws_bedrock:claude-2"
+        guess =
+          case bot_user.id
+          when DiscourseAi::AiBot::EntryPoint::CLAUDE_V2_ID
+            if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-2")
+              "aws_bedrock:claude-2"
+            else
+              "anthropic:claude-2"
+            end
+          when DiscourseAi::AiBot::EntryPoint::GPT4_ID
+            "open_ai:gpt-4"
+          when DiscourseAi::AiBot::EntryPoint::GPT4_TURBO_ID
+            "open_ai:gpt-4-turbo"
+          when DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID
+            "open_ai:gpt-3.5-turbo-16k"
+          when DiscourseAi::AiBot::EntryPoint::MIXTRAL_ID
+            mixtral_model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+            if DiscourseAi::Completions::Endpoints::Vllm.correctly_configured?(mixtral_model)
+              "vllm:#{mixtral_model}"
+            elsif DiscourseAi::Completions::Endpoints::HuggingFace.correctly_configured?(
+                  mixtral_model,
+                )
+              "hugging_face:#{mixtral_model}"
+            else
+              "ollama:mistral"
+            end
+          when DiscourseAi::AiBot::EntryPoint::GEMINI_ID
+            "google:gemini-pro"
+          when DiscourseAi::AiBot::EntryPoint::FAKE_ID
+            "fake:fake"
+          when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_OPUS_ID
+            if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?(
+                 "claude-3-opus",
+               )
+              "aws_bedrock:claude-3-opus"
+            else
+              "anthropic:claude-3-opus"
+            end
+          when DiscourseAi::AiBot::EntryPoint::COHERE_COMMAND_R_PLUS
+            "cohere:command-r-plus"
+          when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_SONNET_ID
+            if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?(
+                 "claude-3-sonnet",
+               )
+              "aws_bedrock:claude-3-sonnet"
+            else
+              "anthropic:claude-3-sonnet"
+            end
+          when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_HAIKU_ID
+            if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?(
+                 "claude-3-haiku",
+               )
+              "aws_bedrock:claude-3-haiku"
+            else
+              "anthropic:claude-3-haiku"
+            end
           else
-            "anthropic:claude-2"
+            nil
           end
-        when DiscourseAi::AiBot::EntryPoint::GPT4_ID
-          "open_ai:gpt-4"
-        when DiscourseAi::AiBot::EntryPoint::GPT4_TURBO_ID
-          "open_ai:gpt-4-turbo"
-        when DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID
-          "open_ai:gpt-3.5-turbo-16k"
-        when DiscourseAi::AiBot::EntryPoint::MIXTRAL_ID
-          mixtral_model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-          if DiscourseAi::Completions::Endpoints::Vllm.correctly_configured?(mixtral_model)
-            "vllm:#{mixtral_model}"
-          elsif DiscourseAi::Completions::Endpoints::HuggingFace.correctly_configured?(
-                mixtral_model,
-              )
-            "hugging_face:#{mixtral_model}"
-          else
-            "ollama:mistral"
-          end
-        when DiscourseAi::AiBot::EntryPoint::GEMINI_ID
-          "google:gemini-pro"
-        when DiscourseAi::AiBot::EntryPoint::FAKE_ID
-          "fake:fake"
-        when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_OPUS_ID
-          if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-3-opus")
-            "aws_bedrock:claude-3-opus"
-          else
-            "anthropic:claude-3-opus"
-          end
-        when DiscourseAi::AiBot::EntryPoint::COHERE_COMMAND_R_PLUS
-          "cohere:command-r-plus"
-        when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_SONNET_ID
-          if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?(
-               "claude-3-sonnet",
-             )
-            "aws_bedrock:claude-3-sonnet"
-          else
-            "anthropic:claude-3-sonnet"
-          end
-        when DiscourseAi::AiBot::EntryPoint::CLAUDE_3_HAIKU_ID
-          if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-3-haiku")
-            "aws_bedrock:claude-3-haiku"
-          else
-            "anthropic:claude-3-haiku"
-          end
-        else
-          nil
+
+        if guess
+          provider, model_name = guess.split(":")
+          llm_model = LlmModel.find_by(provider: provider, name: model_name)
+
+          return "custom:#{llm_model.id}" if llm_model
         end
+
+        guess
       end
 
       def build_placeholder(summary, details, custom_raw: nil)

lib/automation.rb

@@ -25,6 +25,9 @@ module DiscourseAi
     ]
 
     def self.translate_model(model)
+      llm_model = LlmModel.find_by(name: model)
+      return "custom:#{llm_model.id}" if llm_model
+
       return "google:#{model}" if model.start_with? "gemini"
       return "open_ai:#{model}" if model.start_with? "gpt"
       return "cohere:#{model}" if model.start_with? "command"

lib/completions/dialects/chat_gpt.rb

@@ -30,6 +30,8 @@ module DiscourseAi
         end
 
         def max_prompt_tokens
+          return opts[:max_prompt_tokens] if opts.dig(:max_prompt_tokens).present?
+
           # provide a buffer of 120 tokens - our function counting is not
           # 100% accurate and getting numbers to align exactly is very hard
           buffer = (opts[:max_tokens] || 2500) + 50

lib/completions/dialects/claude.rb

@@ -50,6 +50,7 @@ module DiscourseAi
         end
 
         def max_prompt_tokens
+          return opts[:max_prompt_tokens] if opts.dig(:max_prompt_tokens).present?
           # Longer term it will have over 1 million
           200_000 # Claude-3 has a 200k context window for now
         end

lib/completions/dialects/command.rb

@@ -38,6 +38,8 @@ module DiscourseAi
         end
 
         def max_prompt_tokens
+          return opts[:max_prompt_tokens] if opts.dig(:max_prompt_tokens).present?
+
           case model_name
           when "command-light"
             4096
@@ -62,10 +64,6 @@ module DiscourseAi
           self.class.tokenizer.size(context[:content].to_s + context[:name].to_s)
         end
 
-        def tools_dialect
-          @tools_dialect ||= DiscourseAi::Completions::Dialects::XmlTools.new(prompt.tools)
-        end
-
         def system_msg(msg)
           cmd_msg = { role: "SYSTEM", message: msg[:content] }
 

lib/completions/dialects/dialect.rb

@@ -9,14 +9,22 @@ module DiscourseAi
             raise NotImplemented
           end
 
-          def dialect_for(model_name)
-            dialects = [
+          def all_dialects
+            [
               DiscourseAi::Completions::Dialects::ChatGpt,
               DiscourseAi::Completions::Dialects::Gemini,
               DiscourseAi::Completions::Dialects::Mistral,
               DiscourseAi::Completions::Dialects::Claude,
               DiscourseAi::Completions::Dialects::Command,
             ]
+          end
+
+          def available_tokenizers
+            all_dialects.map(&:tokenizer)
+          end
+
+          def dialect_for(model_name)
+            dialects = all_dialects
 
             if Rails.env.test? || Rails.env.development?
               dialects << DiscourseAi::Completions::Dialects::Fake

lib/completions/dialects/gemini.rb

@@ -68,6 +68,8 @@ module DiscourseAi
         end
 
         def max_prompt_tokens
+          return opts[:max_prompt_tokens] if opts.dig(:max_prompt_tokens).present?
+
           if model_name == "gemini-1.5-pro"
             # technically we support 1 million tokens, but we're being conservative
             800_000

lib/completions/dialects/mistral.rb

@@ -23,6 +23,8 @@ module DiscourseAi
         end
 
         def max_prompt_tokens
+          return opts[:max_prompt_tokens] if opts.dig(:max_prompt_tokens).present?
+
           32_000
         end
 

lib/completions/llm.rb

@@ -18,6 +18,14 @@ module DiscourseAi
       UNKNOWN_MODEL = Class.new(StandardError)
 
       class << self
+        def provider_names
+          %w[aws_bedrock anthropic vllm hugging_face cohere open_ai google azure]
+        end
+
+        def tokenizer_names
+          DiscourseAi::Completions::Dialects::Dialect.available_tokenizers.map(&:name).uniq
+        end
+
         def models_by_provider
           # ChatGPT models are listed under open_ai but they are actually available through OpenAI and Azure.
           # However, since they use the same URL/key settings, there's no reason to duplicate them.
@@ -80,36 +88,54 @@ module DiscourseAi
         end
 
         def proxy(model_name)
+          # We are in the process of transitioning to always use objects here.
+          # We'll live with this hack for a while.
           provider_and_model_name = model_name.split(":")
-
           provider_name = provider_and_model_name.first
           model_name_without_prov = provider_and_model_name[1..].join
+          is_custom_model = provider_name == "custom"
+
+          if is_custom_model
+            llm_model = LlmModel.find(model_name_without_prov)
+            provider_name = llm_model.provider
+            model_name_without_prov = llm_model.name
+          end
 
           dialect_klass =
             DiscourseAi::Completions::Dialects::Dialect.dialect_for(model_name_without_prov)
 
+          if is_custom_model
+            tokenizer = llm_model.tokenizer_class
+          else
+            tokenizer = dialect_klass.tokenizer
+          end
+
           if @canned_response
             if @canned_llm && @canned_llm != model_name
               raise "Invalid call LLM call, expected #{@canned_llm} but got #{model_name}"
             end
-            return new(dialect_klass, nil, model_name, gateway: @canned_response)
+            return new(dialect_klass, nil, model_name, opts: { gateway: @canned_response })
           end
 
+          opts = {}
+          opts[:max_prompt_tokens] = llm_model.max_prompt_tokens if is_custom_model
+
           gateway_klass =
             DiscourseAi::Completions::Endpoints::Base.endpoint_for(
               provider_name,
               model_name_without_prov,
             )
 
-          new(dialect_klass, gateway_klass, model_name_without_prov)
+          new(dialect_klass, gateway_klass, model_name_without_prov, opts: opts)
         end
       end
 
-      def initialize(dialect_klass, gateway_klass, model_name, gateway: nil)
+      def initialize(dialect_klass, gateway_klass, model_name, opts: {})
         @dialect_klass = dialect_klass
         @gateway_klass = gateway_klass
         @model_name = model_name
-        @gateway = gateway
+        @gateway = opts[:gateway]
+        @max_prompt_tokens = opts[:max_prompt_tokens]
       end
 
       # @param generic_prompt { DiscourseAi::Completions::Prompt } - Our generic prompt object
@@ -166,11 +192,18 @@ module DiscourseAi
         model_params.keys.each { |key| model_params.delete(key) if model_params[key].nil? }
 
         gateway = @gateway || gateway_klass.new(model_name, dialect_klass.tokenizer)
-        dialect = dialect_klass.new(prompt, model_name, opts: model_params)
+        dialect =
+          dialect_klass.new(
+            prompt,
+            model_name,
+            opts: model_params.merge(max_prompt_tokens: @max_prompt_tokens),
+          )
         gateway.perform_completion!(dialect, user, model_params, &partial_read_blk)
       end
 
       def max_prompt_tokens
+        return @max_prompt_tokens if @max_prompt_tokens.present?
+
         dialect_klass.new(DiscourseAi::Completions::Prompt.new(""), model_name).max_prompt_tokens
       end
 

lib/configuration/llm_enumerator.rb

@@ -10,14 +10,22 @@ module DiscourseAi
       end
 
       def self.values
-        # do not cache cause settings can change this
-        DiscourseAi::Completions::Llm.models_by_provider.flat_map do |provider, models|
-          endpoint =
-            DiscourseAi::Completions::Endpoints::Base.endpoint_for(provider.to_s, models.first)
+        begin
+          llm_models =
+            DiscourseAi::Completions::Llm.models_by_provider.flat_map do |provider, models|
+              endpoint =
+                DiscourseAi::Completions::Endpoints::Base.endpoint_for(provider.to_s, models.first)
 
-          models.map do |model_name|
-            { name: endpoint.display_name(model_name), value: "#{provider}:#{model_name}" }
+              models.map do |model_name|
+                { name: endpoint.display_name(model_name), value: "#{provider}:#{model_name}" }
+              end
+            end
+
+          LlmModel.all.each do |model|
+            llm_models << { name: model.display_name, value: "custom:#{model.id}" }
           end
+
+          llm_models
         end
       end
     end

plugin.rb

@@ -26,6 +26,8 @@ register_asset "stylesheets/modules/sentiment/common/dashboard.scss"
 register_asset "stylesheets/modules/sentiment/desktop/dashboard.scss", :desktop
 register_asset "stylesheets/modules/sentiment/mobile/dashboard.scss", :mobile
 
+register_asset "stylesheets/modules/llms/common/ai-llms-editor.scss"
+
 module ::DiscourseAi
   PLUGIN_NAME = "discourse-ai"
 end

spec/fabricators/llm_model_fabricator.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+Fabricator(:llm_model) do
+  display_name "A good model"
+  name "gpt-4-turbo"
+  provider "open_ai"
+  tokenizer "DiscourseAi::Tokenizers::OpenAi"
+  max_prompt_tokens 32_000
+end

spec/lib/completions/endpoints/hugging_face_spec.rb

@@ -81,7 +81,10 @@ end
 
 RSpec.describe DiscourseAi::Completions::Endpoints::HuggingFace do
   subject(:endpoint) do
-    described_class.new("Llama2-*-chat-hf", DiscourseAi::Tokenizer::Llama2Tokenizer)
+    described_class.new(
+      "mistralai/Mistral-7B-Instruct-v0.2",
+      DiscourseAi::Tokenizer::MixtralTokenizer,
+    )
   end
 
   before { SiteSetting.ai_hugging_face_api_url = "https://test.dev" }

spec/lib/completions/llm_spec.rb

@@ -6,7 +6,9 @@ RSpec.describe DiscourseAi::Completions::Llm do
       DiscourseAi::Completions::Dialects::Mistral,
       canned_response,
       "hugging_face:Upstage-Llama-2-*-instruct-v2",
-      gateway: canned_response,
+      opts: {
+        gateway: canned_response,
+      },
     )
   end
 

spec/requests/admin/ai_llms_controller_spec.rb

@@ -0,0 +1,68 @@
+# frozen_string_literal: true
+
+RSpec.describe DiscourseAi::Admin::AiLlmsController do
+  fab!(:admin)
+
+  before { sign_in(admin) }
+
+  describe "GET #index" do
+    it "includes all available providers metadata" do
+      get "/admin/plugins/discourse-ai/ai-llms.json"
+      expect(response).to be_successful
+
+      expect(response.parsed_body["meta"]["providers"]).to contain_exactly(
+        *DiscourseAi::Completions::Llm.provider_names,
+      )
+    end
+  end
+
+  describe "POST #create" do
+    context "with valid attributes" do
+      let(:valid_attrs) do
+        {
+          display_name: "My cool LLM",
+          name: "gpt-3.5",
+          provider: "open_ai",
+          tokenizer: "DiscourseAi::Tokenizers::OpenAiTokenizer",
+          max_prompt_tokens: 16_000,
+        }
+      end
+
+      it "creates a new LLM model" do
+        post "/admin/plugins/discourse-ai/ai-llms.json", params: { ai_llm: valid_attrs }
+
+        created_model = LlmModel.last
+
+        expect(created_model.display_name).to eq(valid_attrs[:display_name])
+        expect(created_model.name).to eq(valid_attrs[:name])
+        expect(created_model.provider).to eq(valid_attrs[:provider])
+        expect(created_model.tokenizer).to eq(valid_attrs[:tokenizer])
+        expect(created_model.max_prompt_tokens).to eq(valid_attrs[:max_prompt_tokens])
+      end
+    end
+  end
+
+  describe "PUT #update" do
+    fab!(:llm_model)
+
+    context "with valid update params" do
+      let(:update_attrs) { { provider: "anthropic" } }
+
+      it "updates the model" do
+        put "/admin/plugins/discourse-ai/ai-llms/#{llm_model.id}.json",
+            params: {
+              ai_llm: update_attrs,
+            }
+
+        expect(response.status).to eq(200)
+        expect(llm_model.reload.provider).to eq(update_attrs[:provider])
+      end
+
+      it "returns a 404 if there is no model with the given Id" do
+        put "/admin/plugins/discourse-ai/ai-llms/9999999.json"
+
+        expect(response.status).to eq(404)
+      end
+    end
+  end
+end