Handle scenario where vLLM uses a SRV record

app/models/llm_model.rb

@@ -2,9 +2,33 @@
 
 class LlmModel < ActiveRecord::Base
   FIRST_BOT_USER_ID = -1200
+  RESERVED_VLLM_SRV_URL = "https://vllm.shadowed-by-srv.invalid"
 
   belongs_to :user
 
+  validates :url, exclusion: { in: [RESERVED_VLLM_SRV_URL] }
+
+  def self.enable_or_disable_srv_llm!
+    srv_model = find_by(url: RESERVED_VLLM_SRV_URL)
+    if SiteSetting.ai_vllm_endpoint_srv.present? && srv_model.blank?
+      record =
+        new(
+          display_name: "vLLM SRV LLM",
+          name: "mistralai/Mixtral",
+          provider: "vllm",
+          tokenizer: "DiscourseAi::Tokenizer::MixtralTokenizer",
+          url: RESERVED_VLLM_SRV_URL,
+          vllm_key: "",
+          user_id: nil,
+          enabled_chat_bot: false,
+        )
+
+      record.save(validate: false) # Ignore reserved URL validation
+    elsif srv_model.present?
+      srv_model.destroy!
+    end
+  end
+
   def toggle_companion_user
     return if name == "fake" && Rails.env.production?
 

app/serializers/llm_model_serializer.rb

@@ -11,5 +11,10 @@ class LlmModelSerializer < ApplicationSerializer
              :tokenizer,
              :api_key,
              :url,
-             :enabled_chat_bot
+             :enabled_chat_bot,
+             :url_editable
+
+  def url_editable
+    object.url != LlmModel::RESERVED_VLLM_SRV_URL
+  end
 end

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

@@ -131,6 +131,12 @@ export default class AiLlmEditor extends Component {
       @route="adminPlugins.show.discourse-ai-llms"
       @label="discourse_ai.llms.back"
     />
+    {{#unless @model.url_editable}}
+      <div class="alert alert-info">
+        {{icon "exclamation-circle"}}
+        {{I18n.t "discourse_ai.llms.srv_warning"}}
+      </div>
+    {{/unless}}
     <form class="form-horizontal ai-llm-editor">
       <div class="control-group">
         <label>{{i18n "discourse_ai.llms.display_name"}}</label>
@@ -159,14 +165,16 @@ export default class AiLlmEditor extends Component {
           @content={{this.selectedProviders}}
         />
       </div>
-      <div class="control-group">
-        <label>{{I18n.t "discourse_ai.llms.url"}}</label>
-        <Input
-          class="ai-llm-editor-input ai-llm-editor__url"
-          @type="text"
-          @value={{@model.url}}
-        />
-      </div>
+      {{#if @model.url_editable}}
+        <div class="control-group">
+          <label>{{I18n.t "discourse_ai.llms.url"}}</label>
+            <Input
+              class="ai-llm-editor-input ai-llm-editor__url"
+              @type="text"
+              @value={{@model.url}}
+            />
+        </div>
+      {{/if}}
       <div class="control-group">
         <label>{{I18n.t "discourse_ai.llms.api_key"}}</label>
         <Input

config/locales/client.en.yml

@@ -218,6 +218,8 @@ en:
         confirm_delete: Are you sure you want to delete this model?
         delete: Delete
 
+        srv_warning: This LLM points to an SRV record, and its URL is not editable. You have to update the hidden "ai_vllm_endpoint_srv" setting instead.
+
         tests:
           title: "Run Test"
           running: "Running test..."

db/post_migrate/20240528144216_seed_open_ai_models.rb

@@ -5,7 +5,7 @@ def up
     models = []
 
     open_ai_api_key = fetch_setting("ai_openai_api_key")
-    enabled_models = fetch_setting("ai_bot_enabled_chat_bots").to_a.split("|")
+    enabled_models = fetch_setting("ai_bot_enabled_chat_bots")&.split("|").to_a
     enabled_models = ["gpt-3.5-turbo"] if enabled_models.empty?
 
     if open_ai_api_key.present?

db/post_migrate/20240531205234_seed_claude_models.rb

@@ -7,10 +7,10 @@ def up
     models = []
 
     bedrock_secret_access_key = fetch_setting("ai_bedrock_secret_access_key")
-    enabled_models = fetch_setting("ai_bot_enabled_chat_bots").to_a.split("|")
+    enabled_models = fetch_setting("ai_bot_enabled_chat_bots")&.split("|").to_a
 
     if bedrock_secret_access_key.present?
-      bedrock_region = fetch_setting("ai_bedrock_region")
+      bedrock_region = fetch_setting("ai_bedrock_region") || "us-east-1"
 
       claude_models.each do |cm|
         url =

db/post_migrate/20240603133432_seed_other_propietary_models.rb

@@ -5,7 +5,7 @@ def up
     models = []
 
     gemini_key = fetch_setting("ai_gemini_api_key")
-    enabled_models = fetch_setting("ai_bot_enabled_chat_bots").to_a.split("|")
+    enabled_models = fetch_setting("ai_bot_enabled_chat_bots")&.split("|").to_a
 
     if gemini_key.present?
       gemini_models = %w[gemini-pro gemini-1.5-pro gemini-1.5-flash]

db/post_migrate/20240603143158_seed_oss_models.rb

@@ -3,6 +3,8 @@
 class SeedOssModels < ActiveRecord::Migration[7.0]
   def up
     models = []
+    enabled_models = fetch_setting("ai_bot_enabled_chat_bots")&.split("|").to_a
+    enabled = enabled_models.include?("mixtral-8x7B-Instruct-V0.1")
 
     hf_key = fetch_setting("ai_hugging_face_api_key")
     hf_url = fetch_setting("ai_hugging_face_api_url")
@@ -16,7 +18,7 @@ def up
       name = hf_display_name || "mistralai/Mixtral"
       token_limit = hf_token_limit || 32_000
 
-      models << "('#{name}', '#{name}', 'hugging_face', 'DiscourseAi::Tokenizer::MixtralTokenizer', #{token_limit}, '#{hf_url}', '#{hf_key}', #{user_id}, NOW(), NOW())"
+      models << "('#{name}', '#{name}', 'hugging_face', 'DiscourseAi::Tokenizer::MixtralTokenizer', #{token_limit}, '#{hf_url}', '#{hf_key}', #{user_id}, #{enabled}, NOW(), NOW())"
     end
 
     vllm_key = fetch_setting("ai_vllm_api_key")
@@ -26,23 +28,30 @@ def up
       url = "#{vllm_url}/v1/chat/completions"
       name = "mistralai/Mixtral"
 
-      models << "('#{name}', '#{name}', 'vllm', 'DiscourseAi::Tokenizer::MixtralTokenizer', 32000, '#{url}', '#{vllm_key}', #{user_id}, NOW(), NOW())"
+      models << "('#{name}', '#{name}', 'vllm', 'DiscourseAi::Tokenizer::MixtralTokenizer', 32000, '#{url}', '#{vllm_key}', #{user_id}, #{enabled}, NOW(), NOW())"
     end
 
     vllm_srv = fetch_setting("ai_vllm_endpoint_srv")
+    srv_reserved_url = "https://vllm.shadowed-by-srv.invalid"
 
-    if vllm_srv.present? && vllm_key.present?
-      url = "https://shadowed-by-srv.invalid"
+    srv_record =
+      DB.query_single(
+        "SELECT id FROM llm_models WHERE url = :reserved",
+        reserved: srv_reserved_url,
+      ).first
+
+    if vllm_srv.present? && srv.record.nil?
+      url = "https://vllm.shadowed-by-srv.invalid"
       name = "mistralai/Mixtral"
 
-      models << "('#{name}', '#{name}', 'vllm', 'DiscourseAi::Tokenizer::MixtralTokenizer', 32000, '#{url}', '#{vllm_key}', #{user_id}, NOW(), NOW())"
+      models << "('vLLM SRV LLM', '#{name}', 'vllm', 'DiscourseAi::Tokenizer::MixtralTokenizer', 32000, '#{url}', '#{vllm_key}', #{user_id}, #{enabled}, NOW(), NOW())"
     end
 
     if models.present?
       rows = models.compact.join(", ")
 
       DB.exec(<<~SQL, rows: rows) if rows.present?
-        INSERT INTO llm_models(display_name, name, provider, tokenizer, max_prompt_tokens, url, api_key, user_id, created_at, updated_at)
+        INSERT INTO llm_models(display_name, name, provider, tokenizer, max_prompt_tokens, url, api_key, user_id, enabled_chat_bot, created_at, updated_at)
         VALUES #{rows};
       SQL
     end

lib/completions/endpoints/vllm.rb

@@ -44,7 +44,9 @@ def provider_id
         private
 
         def model_uri
-          return URI(llm_model.url) if llm_model&.url
+          if llm_model&.url && !llm_model&.url == LlmModel::RESERVED_VLLM_SRV_URL
+            return URI(llm_model.url)
+          end
 
           service = DiscourseAi::Utils::DnsSrv.lookup(SiteSetting.ai_vllm_endpoint_srv)
           if service.present?

plugin.rb

@@ -43,6 +43,8 @@ module ::DiscourseAi
 
   add_admin_route("discourse_ai.title", "discourse-ai", { use_new_show_route: true })
 
+  LlmModel.enable_or_disable_srv_llm!
+
   [
     DiscourseAi::Embeddings::EntryPoint.new,
     DiscourseAi::Nsfw::EntryPoint.new,