UX: Validations to LLM-backed features (except AI Bot) (#436)

* UX: Validations to Llm-backed features (except AI Bot)

This change is part of an ongoing effort to prevent enabling a broken feature due to lack of configuration. We also want to explicit which provider we are going to use. For example, Claude models are available through AWS Bedrock and Anthropic, but the configuration differs.

Validations are:

* You must choose a model before enabling the feature.
* You must turn off the feature before setting the model to blank.
* You must configure each model settings before being able to select it.

* Add provider name to summarization options

* vLLM can technically support same models as HF

* Check we can talk to the selected model

* Check for Bedrock instead of anthropic as a site could have both creds setup

config/locales/server.en.yml

@@ -240,3 +240,14 @@ en:
           anger: "Anger"
           joy: "Joy"
           disgust: "Disgust"
+
+    llm:
+      configuration:
+        disable_module_first: "You have to disable %{setting} first."
+        set_llm_first: "Set %{setting} first."
+        model_unreachable: "We couldn't get a response from this model. Check your settings first."
+      endpoints:
+        not_configured: "%{display_name} (not configured)"
+        configuration_hint:
+          one: "Make sure the `%{settings}` setting was configured."
+          other: "Make sure these settings were configured: %{settings}"

config/settings.yml

@@ -170,6 +170,7 @@ discourse_ai:
   composer_ai_helper_enabled:
     default: false
     client: true
+    validator: "DiscourseAi::Configuration::LlmDependencyValidator"
   ai_helper_allowed_groups:
     client: true
     type: group_list
@@ -181,17 +182,11 @@ discourse_ai:
     default: false
     client: true
   ai_helper_model:
-    default: gpt-3.5-turbo
+    default: ""
+    allow_any: false
     type: enum
-    choices:
+    enum: "DiscourseAi::Configuration::LlmEnumerator"
-      - gpt-3.5-turbo
+    validator: "DiscourseAi::Configuration::LlmValidator"
-      - gpt-4
-      - claude-2
-      - stable-beluga-2
-      - Llama2-chat-hf
-      - gemini-pro
-      - mistralai/Mixtral-8x7B-Instruct-v0.1
-      - mistralai/Mistral-7B-Instruct-v0.2
   ai_helper_custom_prompts_allowed_groups:
     client: true
     type: group_list
@@ -257,21 +252,13 @@ discourse_ai:
   ai_embeddings_semantic_search_enabled:
     default: false
     client: true
+    validator: "DiscourseAi::Configuration::LlmDependencyValidator"
   ai_embeddings_semantic_search_hyde_model:
-    default: "gpt-3.5-turbo"
+    default: ""
     type: enum
     allow_any: false
-    choices:
+    enum: "DiscourseAi::Configuration::LlmEnumerator"
-      - Llama2-*-chat-hf
+    validator: "DiscourseAi::Configuration::LlmValidator"
-      - claude-instant-1
-      - claude-2
-      - gpt-3.5-turbo
-      - gpt-4
-      - StableBeluga2
-      - Upstage-Llama-2-*-instruct-v2
-      - gemini-pro
-      - mistralai/Mixtral-8x7B-Instruct-v0.1
-      - mistralai/Mistral-7B-Instruct-v0.2
 
   ai_summarization_discourse_service_api_endpoint: ""
   ai_summarization_discourse_service_api_endpoint_srv:

db/post_migrate/20240119152348_explicit_provider_backwards_compat.rb

@@ -0,0 +1,91 @@
+# frozen_string_literal: true
+
+class ExplicitProviderBackwardsCompat < ActiveRecord::Migration[7.0]
+  def up
+    backfill_settings("composer_ai_helper_enabled", "ai_helper_model")
+    backfill_settings(
+      "ai_embeddings_semantic_search_enabled",
+      "ai_embeddings_semantic_search_hyde_model",
+    )
+  end
+
+  def down
+    raise ActiveRecord::IrreversibleMigration
+  end
+
+  def backfill_settings(feature_setting_name, llm_setting_name)
+    feature_enabled =
+      DB.query_single(
+        "SELECT value FROM site_settings WHERE name = :setting_name",
+        setting_name: feature_setting_name,
+      ).first == "t"
+
+    setting_value =
+      DB
+        .query_single(
+          "SELECT value FROM site_settings WHERE name = :llm_setting",
+          llm_setting: llm_setting_name,
+        )
+        .first
+        .to_s
+
+    providers = %w[aws_bedrock anthropic open_ai hugging_face vllm google]
+    # Sanity check to make sure we won't add provider twice.
+    return if providers.include?(setting_value.split(":").first)
+
+    if !setting_value && feature_enabled
+      # Enabled and using old default (gpt-3.5-turbo)
+      DB.exec(
+        "UPDATE site_settings SET value='open_ai:gpt-3.5-turbo' WHERE name=:llm_setting",
+        llm_setting: llm_setting_name,
+      )
+    elsif setting_value && !feature_enabled
+      # They'll have to choose an LLM model again before enabling the feature
+      DB.exec("DELETE FROM site_settings WHERE name=:llm_setting", llm_setting: llm_setting_name)
+    elsif setting_value && feature_enabled
+      DB.exec(
+        "UPDATE site_settings SET value=:new_value WHERE name=:llm_setting",
+        llm_setting: llm_setting_name,
+        new_value: append_provider(setting_value),
+      )
+    end
+  end
+
+  def append_provider(value)
+    open_ai_models = %w[gpt-3.5-turbo gpt-4 gpt-3.5-turbo-16k gpt-4-32k gpt-4-turbo]
+    return "open_ai:#{value}" if open_ai_models.include?(value)
+    return "google:#{value}" if value == "gemini-pro"
+
+    hf_models = %w[StableBeluga2 Upstage-Llama-2-*-instruct-v2 Llama2-*-chat-hf Llama2-chat-hf]
+    return "hugging_face:#{value}" if hf_models.include?(value)
+
+    # Models available through multiple providers
+    claude_models = %w[claude-instant-1 claude-2]
+    if claude_models.include?(value)
+      has_bedrock_creds =
+        DB.query_single(
+          "SELECT value FROM site_settings WHERE name = 'ai_bedrock_secret_access_key' OR name = 'ai_bedrock_access_key_id' ",
+        ).length > 0
+
+      if has_bedrock_creds
+        return "aws_bedrock:#{value}"
+      else
+        return "anthropic:#{value}"
+      end
+    end
+
+    mixtral_models = %w[mistralai/Mixtral-8x7B-Instruct-v0.1 mistralai/Mistral-7B-Instruct-v0.2]
+    if mixtral_models.include?(value)
+      vllm_configured =
+        DB.query_single(
+          "SELECT value FROM site_settings WHERE name = 'ai_vllm_endpoint_srv' OR name = 'ai_vllm_endpoint' ",
+        ).length > 0
+
+      if vllm_configured
+        "vllm:#{value}"
+      else
+        "hugging_face:#{value}"
+      end
+    end
+  end
+end

lib/ai_bot/bot.rb

@@ -125,28 +125,39 @@ module DiscourseAi
       end
 
       def model(prefer_low_cost: false)
+        # HACK(roman): We'll do this until we define how we represent different providers in the bot settings
         default_model =
           case bot_user.id
           when DiscourseAi::AiBot::EntryPoint::CLAUDE_V2_ID
-            "claude-2"
+            if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-2")
+              "aws_bedrock:claude-2"
+            else
+              "anthropic:claude-2"
+            end
           when DiscourseAi::AiBot::EntryPoint::GPT4_ID
-            "gpt-4"
+            "open_ai:gpt-4"
           when DiscourseAi::AiBot::EntryPoint::GPT4_TURBO_ID
-            "gpt-4-turbo"
+            "open_ai:gpt-4-turbo"
           when DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID
-            "gpt-3.5-turbo-16k"
+            "open_ai:gpt-3.5-turbo-16k"
           when DiscourseAi::AiBot::EntryPoint::MIXTRAL_ID
-            "mistralai/Mixtral-8x7B-Instruct-v0.1"
+            if DiscourseAi::Completions::Endpoints::Vllm.correctly_configured?(
+                 "mistralai/Mixtral-8x7B-Instruct-v0.1",
+               )
+              "vllm:mistralai/Mixtral-8x7B-Instruct-v0.1"
+            else
+              "hugging_face:mistralai/Mixtral-8x7B-Instruct-v0.1"
+            end
           when DiscourseAi::AiBot::EntryPoint::GEMINI_ID
-            "gemini-pro"
+            "google:gemini-pro"
           when DiscourseAi::AiBot::EntryPoint::FAKE_ID
-            "fake"
+            "fake:fake"
           else
             nil
           end
 
-        if %w[gpt-4 gpt-4-turbo].include?(default_model) && prefer_low_cost
+        if %w[open_ai:gpt-4 open_ai:gpt-4-turbo].include?(default_model) && prefer_low_cost
-          return "gpt-3.5-turbo-16k"
+          return "open_ai:gpt-3.5-turbo-16k"
         end
 
         default_model

lib/completions/dialects/fake.rb

@@ -9,6 +9,10 @@ module DiscourseAi
             model_name == "fake"
           end
 
+          def translate
+            ""
+          end
+
           def tokenizer
             DiscourseAi::Tokenizer::OpenAiTokenizer
           end

lib/completions/endpoints/anthropic.rb

@@ -4,8 +4,22 @@ module DiscourseAi
   module Completions
     module Endpoints
       class Anthropic < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[claude-instant-1 claude-2].include?(model_name)
+          def can_contact?(endpoint_name, model_name)
+            endpoint_name == "anthropic" && %w[claude-instant-1 claude-2].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[ai_anthropic_api_key]
+          end
+
+          def correctly_configured?(_model_name)
+            SiteSetting.ai_anthropic_api_key.present?
+          end
+
+          def endpoint_name(model_name)
+            "Anthropic - #{model_name}"
+          end
         end
 
         def normalize_model_params(model_params)

lib/completions/endpoints/aws_bedrock.rb

@@ -6,11 +6,24 @@ module DiscourseAi
   module Completions
     module Endpoints
       class AwsBedrock < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[claude-instant-1 claude-2].include?(model_name) &&
+          def can_contact?(endpoint_name, model_name)
+            endpoint_name == "aws_bedrock" && %w[claude-instant-1 claude-2].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[ai_bedrock_access_key_id ai_bedrock_secret_access_key ai_bedrock_region]
+          end
+
+          def correctly_configured?(_model_name)
             SiteSetting.ai_bedrock_access_key_id.present? &&
-            SiteSetting.ai_bedrock_secret_access_key.present? &&
+              SiteSetting.ai_bedrock_secret_access_key.present? &&
-            SiteSetting.ai_bedrock_region.present?
+              SiteSetting.ai_bedrock_region.present?
+          end
+
+          def endpoint_name(model_name)
+            "AWS Bedrock - #{model_name}"
+          end
         end
 
         def normalize_model_params(model_params)

lib/completions/endpoints/base.rb

@@ -7,29 +7,54 @@ module DiscourseAi
         CompletionFailed = Class.new(StandardError)
         TIMEOUT = 60
 
-        def self.endpoint_for(model_name)
+        class << self
-          # Order is important.
+          def endpoint_for(provider_name, model_name)
-          # Bedrock has priority over Anthropic if credentials are present.
+            endpoints = [
-          endpoints = [
+              DiscourseAi::Completions::Endpoints::AwsBedrock,
-            DiscourseAi::Completions::Endpoints::AwsBedrock,
+              DiscourseAi::Completions::Endpoints::Anthropic,
-            DiscourseAi::Completions::Endpoints::Anthropic,
+              DiscourseAi::Completions::Endpoints::OpenAi,
-            DiscourseAi::Completions::Endpoints::OpenAi,
+              DiscourseAi::Completions::Endpoints::HuggingFace,
-            DiscourseAi::Completions::Endpoints::HuggingFace,
+              DiscourseAi::Completions::Endpoints::Gemini,
-            DiscourseAi::Completions::Endpoints::Gemini,
+              DiscourseAi::Completions::Endpoints::Vllm,
-            DiscourseAi::Completions::Endpoints::Vllm,
+            ]
-          ]
 
-          if Rails.env.test? || Rails.env.development?
+            if Rails.env.test? || Rails.env.development?
-            endpoints << DiscourseAi::Completions::Endpoints::Fake
+              endpoints << DiscourseAi::Completions::Endpoints::Fake
+            end
+
+            endpoints.detect(-> { raise DiscourseAi::Completions::Llm::UNKNOWN_MODEL }) do |ek|
+              ek.can_contact?(provider_name, model_name)
+            end
           end
 
-          endpoints.detect(-> { raise DiscourseAi::Completions::Llm::UNKNOWN_MODEL }) do |ek|
+          def configuration_hint
-            ek.can_contact?(model_name)
+            settings = dependant_setting_names
+            I18n.t(
+              "discourse_ai.llm.endpoints.configuration_hint",
+              settings: settings.join(", "),
+              count: settings.length,
+            )
           end
-        end
 
-        def self.can_contact?(_model_name)
+          def display_name(model_name)
-          raise NotImplementedError
+            to_display = endpoint_name(model_name)
+
+            return to_display if correctly_configured?(model_name)
+
+            I18n.t("discourse_ai.llm.endpoints.not_configured", display_name: to_display)
+          end
+
+          def dependant_setting_names
+            raise NotImplementedError
+          end
+
+          def endpoint_name(_model_name)
+            raise NotImplementedError
+          end
+
+          def can_contact?(_endpoint_name, _model_name)
+            raise NotImplementedError
+          end
         end
 
         def initialize(model_name, tokenizer)

lib/completions/endpoints/canned_response.rb

@@ -6,7 +6,7 @@ module DiscourseAi
       class CannedResponse
         CANNED_RESPONSE_ERROR = Class.new(StandardError)
 
-        def self.can_contact?(_)
+        def self.can_contact?(_, _)
           Rails.env.test?
         end
 

lib/completions/endpoints/fake.rb

@@ -4,8 +4,18 @@ module DiscourseAi
   module Completions
     module Endpoints
       class Fake < Base
-        def self.can_contact?(model_name)
+        class << self
-          model_name == "fake"
+          def can_contact?(_endpoint_name, model_name)
+            model_name == "fake"
+          end
+
+          def correctly_configured?(_model_name)
+            true
+          end
+
+          def endpoint_name(_model_name)
+            "Test - fake model"
+          end
         end
 
         STOCK_CONTENT = <<~TEXT

lib/completions/endpoints/gemini.rb

@@ -4,8 +4,23 @@ module DiscourseAi
   module Completions
     module Endpoints
       class Gemini < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[gemini-pro].include?(model_name)
+          def can_contact?(endpoint_name, model_name)
+            return false unless endpoint_name == "google"
+            %w[gemini-pro].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[ai_gemini_api_key]
+          end
+
+          def correctly_configured?(_model_name)
+            SiteSetting.ai_gemini_api_key.present?
+          end
+
+          def endpoint_name(model_name)
+            "Google - #{model_name}"
+          end
         end
 
         def default_options

lib/completions/endpoints/hugging_face.rb

@@ -4,15 +4,31 @@ module DiscourseAi
   module Completions
     module Endpoints
       class HuggingFace < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[
+          def can_contact?(endpoint_name, model_name)
-            StableBeluga2
+            return false unless endpoint_name == "hugging_face"
-            Upstage-Llama-2-*-instruct-v2
+
-            Llama2-*-chat-hf
+            %w[
-            Llama2-chat-hf
+              StableBeluga2
-            mistralai/Mixtral-8x7B-Instruct-v0.1
+              Upstage-Llama-2-*-instruct-v2
-            mistralai/Mistral-7B-Instruct-v0.2
+              Llama2-*-chat-hf
-          ].include?(model_name) && SiteSetting.ai_hugging_face_api_url.present?
+              Llama2-chat-hf
+              mistralai/Mixtral-8x7B-Instruct-v0.1
+              mistralai/Mistral-7B-Instruct-v0.2
+            ].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[ai_hugging_face_api_url]
+          end
+
+          def correctly_configured?(_model_name)
+            SiteSetting.ai_hugging_face_api_url.present?
+          end
+
+          def endpoint_name(model_name)
+            "Hugging Face - #{model_name}"
+          end
         end
 
         def default_options

lib/completions/endpoints/open_ai.rb

@@ -4,15 +4,62 @@ module DiscourseAi
   module Completions
     module Endpoints
       class OpenAi < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[
+          def can_contact?(endpoint_name, model_name)
-            gpt-3.5-turbo
+            return false unless endpoint_name == "open_ai"
-            gpt-4
+
-            gpt-3.5-turbo-16k
+            %w[
-            gpt-4-32k
+              gpt-3.5-turbo
-            gpt-4-0125-preview
+              gpt-4
-            gpt-4-turbo
+              gpt-3.5-turbo-16k
-          ].include?(model_name)
+              gpt-4-32k
+              gpt-4-0125-preview
+              gpt-4-turbo
+            ].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[
+              ai_openai_api_key
+              ai_openai_gpt4_32k_url
+              ai_openai_gpt4_turbo_url
+              ai_openai_gpt4_url
+              ai_openai_gpt4_url
+              ai_openai_gpt35_16k_url
+              ai_openai_gpt35_url
+            ]
+          end
+
+          def correctly_configured?(model_name)
+            SiteSetting.ai_openai_api_key.present? && has_url?(model_name)
+          end
+
+          def has_url?(model)
+            url =
+              if model.include?("gpt-4")
+                if model.include?("32k")
+                  SiteSetting.ai_openai_gpt4_32k_url
+                else
+                  if model.include?("1106") || model.include?("turbo")
+                    SiteSetting.ai_openai_gpt4_turbo_url
+                  else
+                    SiteSetting.ai_openai_gpt4_url
+                  end
+                end
+              else
+                if model.include?("16k")
+                  SiteSetting.ai_openai_gpt35_16k_url
+                else
+                  SiteSetting.ai_openai_gpt35_url
+                end
+              end
+
+            url.present?
+          end
+
+          def endpoint_name(model_name)
+            "OpenAI - #{model_name}"
+          end
         end
 
         def normalize_model_params(model_params)

lib/completions/endpoints/vllm.rb

@@ -4,10 +4,30 @@ module DiscourseAi
   module Completions
     module Endpoints
       class Vllm < Base
-        def self.can_contact?(model_name)
+        class << self
-          %w[mistralai/Mixtral-8x7B-Instruct-v0.1 mistralai/Mistral-7B-Instruct-v0.2].include?(
+          def can_contact?(endpoint_name, model_name)
-            model_name,
+            endpoint_name == "vllm" &&
-          )
+              %w[
+                mistralai/Mixtral-8x7B-Instruct-v0.1
+                mistralai/Mistral-7B-Instruct-v0.2
+                StableBeluga2
+                Upstage-Llama-2-*-instruct-v2
+                Llama2-*-chat-hf
+                Llama2-chat-hf
+              ].include?(model_name)
+          end
+
+          def dependant_setting_names
+            %w[ai_vllm_endpoint_srv ai_vllm_endpoint]
+          end
+
+          def correctly_configured?(_model_name)
+            SiteSetting.ai_vllm_endpoint_srv.present? || SiteSetting.ai_vllm_endpoint.present?
+          end
+
+          def endpoint_name(model_name)
+            "vLLM - #{model_name}"
+          end
         end
 
         def normalize_model_params(model_params)

lib/completions/llm.rb

@@ -7,7 +7,7 @@
 # the target model and routes the completion request through the correct gateway.
 #
 # Use the .proxy method to instantiate an object.
-# It chooses the best dialect and endpoint for the model you want to interact with.
+# It chooses the correct dialect and endpoint for the model you want to interact with.
 #
 # Tests of modules that perform LLM calls can use .with_prepared_responses to return canned responses
 # instead of relying on WebMock stubs like we did in the past.
@@ -17,27 +17,62 @@ module DiscourseAi
     class Llm
       UNKNOWN_MODEL = Class.new(StandardError)
 
-      def self.with_prepared_responses(responses)
+      class << self
-        @canned_response = DiscourseAi::Completions::Endpoints::CannedResponse.new(responses)
+        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.
+          {
+            aws_bedrock: %w[claude-instant-1 claude-2],
+            anthropic: %w[claude-instant-1 claude-2],
+            vllm: %w[
+              mistralai/Mixtral-8x7B-Instruct-v0.1
+              mistralai/Mistral-7B-Instruct-v0.2
+              StableBeluga2
+              Upstage-Llama-2-*-instruct-v2
+              Llama2-*-chat-hf
+              Llama2-chat-hf
+            ],
+            hugging_face: %w[
+              mistralai/Mixtral-8x7B-Instruct-v0.1
+              mistralai/Mistral-7B-Instruct-v0.2
+              StableBeluga2
+              Upstage-Llama-2-*-instruct-v2
+              Llama2-*-chat-hf
+              Llama2-chat-hf
+            ],
+            open_ai: %w[gpt-3.5-turbo gpt-4 gpt-3.5-turbo-16k gpt-4-32k gpt-4-turbo],
+            google: %w[gemini-pro],
+          }.tap { |h| h[:fake] = ["fake"] if Rails.env.test? || Rails.env.development? }
+        end
 
-        yield(@canned_response)
+        def with_prepared_responses(responses)
-      ensure
+          @canned_response = DiscourseAi::Completions::Endpoints::CannedResponse.new(responses)
-        # Don't leak prepared response if there's an exception.
-        @canned_response = nil
-      end
 
-      def self.proxy(model_name)
+          yield(@canned_response)
-        dialect_klass = DiscourseAi::Completions::Dialects::Dialect.dialect_for(model_name)
+        ensure
+          # Don't leak prepared response if there's an exception.
+          @canned_response = nil
+        end
 
-        return new(dialect_klass, @canned_response, model_name) if @canned_response
+        def proxy(model_name)
+          provider_and_model_name = model_name.split(":")
 
-        gateway =
+          provider_name = provider_and_model_name.first
-          DiscourseAi::Completions::Endpoints::Base.endpoint_for(model_name).new(
+          model_name_without_prov = provider_and_model_name[1..].join
-            model_name,
-            dialect_klass.tokenizer,
-          )
 
-        new(dialect_klass, gateway, model_name)
+          dialect_klass =
+            DiscourseAi::Completions::Dialects::Dialect.dialect_for(model_name_without_prov)
+
+          return new(dialect_klass, @canned_response, model_name) if @canned_response
+
+          gateway =
+            DiscourseAi::Completions::Endpoints::Base.endpoint_for(
+              provider_name,
+              model_name_without_prov,
+            ).new(model_name_without_prov, dialect_klass.tokenizer)
+
+          new(dialect_klass, gateway, model_name_without_prov)
+        end
       end
 
       def initialize(dialect_klass, gateway, model_name)

lib/configuration/llm_dependency_validator.rb

@@ -0,0 +1,29 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module Configuration
+    class LlmDependencyValidator
+      def initialize(opts = {})
+        @opts = opts
+      end
+
+      def valid_value?(val)
+        return true if val == "f"
+
+        SiteSetting.public_send(llm_dependency_setting_name).present?
+      end
+
+      def error_message
+        I18n.t("discourse_ai.llm.configuration.set_llm_first", setting: llm_dependency_setting_name)
+      end
+
+      def llm_dependency_setting_name
+        if @opts[:name] == :ai_embeddings_semantic_search_enabled
+          :ai_embeddings_semantic_search_hyde_model
+        else
+          :ai_helper_model
+        end
+      end
+    end
+  end
+end

lib/configuration/llm_enumerator.rb

@@ -0,0 +1,25 @@
+# frozen_string_literal: true
+
+require "enum_site_setting"
+
+module DiscourseAi
+  module Configuration
+    class LlmEnumerator < ::EnumSiteSetting
+      def self.valid_value?(val)
+        true
+      end
+
+      def self.values
+        @values ||=
+          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}" }
+            end
+          end
+      end
+    end
+  end
+end

lib/configuration/llm_validator.rb

@@ -0,0 +1,77 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module Configuration
+    class LlmValidator
+      def initialize(opts = {})
+        @opts = opts
+      end
+
+      def valid_value?(val)
+        if val == ""
+          @parent_enabled = SiteSetting.public_send(parent_module_name)
+          return !@parent_enabled
+        end
+
+        provider_and_model_name = val.split(":")
+
+        provider_name = provider_and_model_name.first
+        model_name_without_prov = provider_and_model_name[1..].join
+
+        endpoint =
+          DiscourseAi::Completions::Endpoints::Base.endpoint_for(
+            provider_name,
+            model_name_without_prov,
+          )
+
+        return false if endpoint.nil?
+
+        if !endpoint.correctly_configured?(model_name_without_prov)
+          @endpoint = endpoint
+          return false
+        end
+
+        if !can_talk_to_model?(val)
+          @unreachable = true
+          return false
+        end
+
+        true
+      end
+
+      def error_message
+        if @parent_enabled
+          return(
+            I18n.t(
+              "discourse_ai.llm.configuration.disable_module_first",
+              setting: parent_module_name,
+            )
+          )
+        end
+
+        return(I18n.t("discourse_ai.llm.configuration.model_unreachable")) if @unreachable
+
+        @endpoint&.configuration_hint
+      end
+
+      def parent_module_name
+        if @opts[:name] == :ai_embeddings_semantic_search_hyde_model
+          :ai_embeddings_semantic_search_enabled
+        else
+          :composer_ai_helper_enabled
+        end
+      end
+
+      private
+
+      def can_talk_to_model?(model_name)
+        DiscourseAi::Completions::Llm
+          .proxy(model_name)
+          .generate("How much is 1 + 1?", user: nil)
+          .present?
+      rescue StandardError
+        false
+      end
+    end
+  end
+end

lib/summarization/entry_point.rb

@@ -5,22 +5,45 @@ module DiscourseAi
     class EntryPoint
       def inject_into(plugin)
         foldable_models = [
-          Models::OpenAi.new("gpt-4", max_tokens: 8192),
+          Models::OpenAi.new("open_ai:gpt-4", max_tokens: 8192),
-          Models::OpenAi.new("gpt-4-32k", max_tokens: 32_768),
+          Models::OpenAi.new("open_ai:gpt-4-32k", max_tokens: 32_768),
-          Models::OpenAi.new("gpt-4-0125-preview", max_tokens: 100_000),
+          Models::OpenAi.new("open_ai:gpt-4-0125-preview", max_tokens: 100_000),
-          Models::OpenAi.new("gpt-3.5-turbo", max_tokens: 4096),
+          Models::OpenAi.new("open_ai:gpt-3.5-turbo", max_tokens: 4096),
-          Models::OpenAi.new("gpt-3.5-turbo-16k", max_tokens: 16_384),
+          Models::OpenAi.new("open_ai:gpt-3.5-turbo-16k", max_tokens: 16_384),
-          Models::Anthropic.new("claude-2", max_tokens: 200_000),
+          Models::Llama2.new(
-          Models::Anthropic.new("claude-instant-1", max_tokens: 100_000),
+            "hugging_face:Llama2-chat-hf",
-          Models::Llama2.new("Llama2-chat-hf", max_tokens: SiteSetting.ai_hugging_face_token_limit),
-          Models::Llama2FineTunedOrcaStyle.new(
-            "StableBeluga2",
             max_tokens: SiteSetting.ai_hugging_face_token_limit,
           ),
-          Models::Gemini.new("gemini-pro", max_tokens: 32_768),
+          Models::Llama2FineTunedOrcaStyle.new(
-          Models::Mixtral.new("mistralai/Mixtral-8x7B-Instruct-v0.1", max_tokens: 32_000),
+            "hugging_face:StableBeluga2",
+            max_tokens: SiteSetting.ai_hugging_face_token_limit,
+          ),
+          Models::Gemini.new("google:gemini-pro", max_tokens: 32_768),
         ]
 
+        claude_prov = "anthropic"
+        if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-2")
+          claude_prov = "aws_bedrock"
+        end
+
+        foldable_models << Models::Anthropic.new("#{claude_prov}:claude-2", max_tokens: 200_000)
+        foldable_models << Models::Anthropic.new(
+          "#{claude_prov}:claude-instant-1",
+          max_tokens: 100_000,
+        )
+
+        mixtral_prov = "hugging_face"
+        if DiscourseAi::Completions::Endpoints::Vllm.correctly_configured?(
+             "mistralai/Mixtral-8x7B-Instruct-v0.1",
+           )
+          mixtral_prov = "vllm"
+        end
+
+        foldable_models << Models::Mixtral.new(
+          "#{mixtral_prov}:mistralai/Mixtral-8x7B-Instruct-v0.1",
+          max_tokens: 32_000,
+        )
+
         foldable_models.each do |model|
           plugin.register_summarization_strategy(Strategies::FoldContent.new(model))
         end

lib/summarization/models/base.rb

@@ -4,8 +4,8 @@ module DiscourseAi
   module Summarization
     module Models
       class Base
-        def initialize(model, max_tokens:)
+        def initialize(model_name, max_tokens:)
-          @model = model
+          @model_name = model_name
           @max_tokens = max_tokens
         end
 
@@ -25,7 +25,11 @@ module DiscourseAi
           max_tokens - reserved_tokens
         end
 
-        attr_reader :model, :max_tokens
+        def model
+          model_name.split(":").last
+        end
+
+        attr_reader :model_name, :max_tokens
 
         protected
 

lib/summarization/strategies/fold_content.rb

@@ -19,7 +19,7 @@ module DiscourseAi
         def summarize(content, user, &on_partial_blk)
           opts = content.except(:contents)
 
-          llm = DiscourseAi::Completions::Llm.proxy(completion_model.model)
+          llm = DiscourseAi::Completions::Llm.proxy(completion_model.model_name)
 
           initial_chunks =
             rebalance_chunks(

spec/jobs/regular/stream_post_helper_spec.rb

@@ -3,6 +3,8 @@
 RSpec.describe Jobs::StreamPostHelper do
   subject(:job) { described_class.new }
 
+  before { SiteSetting.ai_helper_model = "fake:fake" }
+
   describe "#execute" do
     fab!(:topic) { Fabricate(:topic) }
     fab!(:post) do

spec/lib/completions/llm_spec.rb

@@ -5,7 +5,7 @@ RSpec.describe DiscourseAi::Completions::Llm do
     described_class.new(
       DiscourseAi::Completions::Dialects::OrcaStyle,
       canned_response,
-      "Upstage-Llama-2-*-instruct-v2",
+      "hugging_face:Upstage-Llama-2-*-instruct-v2",
     )
   end
 
@@ -13,7 +13,7 @@ RSpec.describe DiscourseAi::Completions::Llm do
 
   describe ".proxy" do
     it "raises an exception when we can't proxy the model" do
-      fake_model = "unknown_v2"
+      fake_model = "unknown:unknown_v2"
 
       expect { described_class.proxy(fake_model) }.to(
         raise_error(DiscourseAi::Completions::Llm::UNKNOWN_MODEL),
@@ -27,7 +27,7 @@ RSpec.describe DiscourseAi::Completions::Llm do
       DiscourseAi::Completions::Endpoints::Fake.chunk_count = 10
     end
 
-    let(:llm) { described_class.proxy("fake") }
+    let(:llm) { described_class.proxy("fake:fake") }
 
     let(:prompt) do
       DiscourseAi::Completions::Prompt.new(

spec/lib/modules/ai_bot/tools/dall_e_spec.rb

@@ -6,7 +6,7 @@ RSpec.describe DiscourseAi::AiBot::Tools::DallE do
   let(:prompts) { ["a pink cow", "a red cow"] }
 
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
   let(:progress_blk) { Proc.new {} }
 
   before { SiteSetting.ai_bot_enabled = true }

spec/lib/modules/ai_bot/tools/db_schema_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::DbSchema do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before { SiteSetting.ai_bot_enabled = true }
   describe "#process" do

spec/lib/modules/ai_bot/tools/google_spec.rb

@@ -4,7 +4,7 @@ RSpec.describe DiscourseAi::AiBot::Tools::Google do
   subject(:search) { described_class.new({ query: "some search term" }) }
 
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
   let(:progress_blk) { Proc.new {} }
 
   before { SiteSetting.ai_bot_enabled = true }

spec/lib/modules/ai_bot/tools/image_spec.rb

@@ -3,7 +3,7 @@
 RSpec.describe DiscourseAi::AiBot::Tools::Image do
   subject(:tool) { described_class.new({ prompts: prompts, seeds: [99, 32] }) }
 
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
   let(:progress_blk) { Proc.new {} }
 
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }

spec/lib/modules/ai_bot/tools/list_categories_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::ListCategories do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before { SiteSetting.ai_bot_enabled = true }
 

spec/lib/modules/ai_bot/tools/list_tags_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::ListTags do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before do
     SiteSetting.ai_bot_enabled = true

spec/lib/modules/ai_bot/tools/read_spec.rb

@@ -4,7 +4,7 @@ RSpec.describe DiscourseAi::AiBot::Tools::Read do
   subject(:tool) { described_class.new({ topic_id: topic_with_tags.id }) }
 
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   fab!(:parent_category) { Fabricate(:category, name: "animals") }
   fab!(:category) { Fabricate(:category, parent_category: parent_category, name: "amazing-cat") }

spec/lib/modules/ai_bot/tools/search_settings_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::SearchSettings do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before { SiteSetting.ai_bot_enabled = true }
 

spec/lib/modules/ai_bot/tools/search_spec.rb

@@ -4,8 +4,10 @@ RSpec.describe DiscourseAi::AiBot::Tools::Search do
   before { SearchIndexer.enable }
   after { SearchIndexer.disable }
 
+  before { SiteSetting.ai_openai_api_key = "asd" }
+
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
   let(:progress_blk) { Proc.new {} }
 
   fab!(:admin)
@@ -65,6 +67,7 @@ RSpec.describe DiscourseAi::AiBot::Tools::Search do
       after { DiscourseAi::Embeddings::SemanticSearch.clear_cache_for(query) }
 
       it "supports semantic search when enabled" do
+        SiteSetting.ai_embeddings_semantic_search_hyde_model = "fake:fake"
         SiteSetting.ai_embeddings_semantic_search_enabled = true
         SiteSetting.ai_embeddings_discourse_service_api_endpoint = "http://test.com"
 

spec/lib/modules/ai_bot/tools/setting_context_spec.rb

@@ -10,7 +10,7 @@ end
 
 RSpec.describe DiscourseAi::AiBot::Tools::SettingContext, if: has_rg? do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before { SiteSetting.ai_bot_enabled = true }
 

spec/lib/modules/ai_bot/tools/summarize_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::Summarize do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
   let(:progress_blk) { Proc.new {} }
 
   before { SiteSetting.ai_bot_enabled = true }

spec/lib/modules/ai_bot/tools/time_spec.rb

@@ -2,7 +2,7 @@
 
 RSpec.describe DiscourseAi::AiBot::Tools::Time do
   let(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
-  let(:llm) { DiscourseAi::Completions::Llm.proxy("gpt-3.5-turbo") }
+  let(:llm) { DiscourseAi::Completions::Llm.proxy("open_ai:gpt-3.5-turbo") }
 
   before { SiteSetting.ai_bot_enabled = true }
 

spec/lib/modules/ai_helper/assistant_spec.rb

@@ -4,6 +4,8 @@ RSpec.describe DiscourseAi::AiHelper::Assistant do
   fab!(:user) { Fabricate(:user) }
   let(:prompt) { CompletionPrompt.find_by(id: mode) }
 
+  before { SiteSetting.ai_helper_model = "fake:fake" }
+
   let(:english_text) { <<~STRING }
     To perfect his horror, Caesar, surrounded at the base of the statue by the impatient daggers of his friends,
     discovers among the faces and blades that of Marcus Brutus, his protege, perhaps his son, and he no longer

spec/lib/modules/ai_helper/chat_thread_titler_spec.rb

@@ -3,6 +3,8 @@
 RSpec.describe DiscourseAi::AiHelper::ChatThreadTitler do
   subject(:titler) { described_class.new(thread) }
 
+  before { SiteSetting.ai_helper_model = "fake:fake" }
+
   fab!(:thread) { Fabricate(:chat_thread) }
   fab!(:user) { Fabricate(:user) }
 

spec/lib/modules/ai_helper/painter_spec.rb

@@ -6,6 +6,7 @@ RSpec.describe DiscourseAi::AiHelper::Painter do
   fab!(:user) { Fabricate(:user) }
 
   before do
+    SiteSetting.ai_helper_model = "fake:fake"
     SiteSetting.ai_stability_api_url = "https://api.stability.dev"
     SiteSetting.ai_stability_api_key = "abc"
     SiteSetting.ai_openai_api_key = "abc"

spec/lib/modules/automation/llm_triage_spec.rb

@@ -10,7 +10,7 @@ describe DiscourseAi::Automation::LlmTriage do
     DiscourseAi::Completions::Llm.with_prepared_responses(["good"]) do
       triage(
         post: post,
-        model: "gpt-4",
+        model: "fake:fake",
         hide_topic: true,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
@@ -20,25 +20,11 @@ describe DiscourseAi::Automation::LlmTriage do
     expect(post.topic.reload.visible).to eq(true)
   end
 
-  it "can hide topics on triage with claude" do
+  it "can hide topics on triage" do
     DiscourseAi::Completions::Llm.with_prepared_responses(["bad"]) do
       triage(
         post: post,
-        model: "claude-2",
+        model: "fake:fake",
-        hide_topic: true,
-        system_prompt: "test %%POST%%",
-        search_for_text: "bad",
-      )
-    end
-
-    expect(post.topic.reload.visible).to eq(false)
-  end
-
-  it "can hide topics on triage with claude" do
-    DiscourseAi::Completions::Llm.with_prepared_responses(["bad"]) do
-      triage(
-        post: post,
-        model: "gpt-4",
         hide_topic: true,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
@@ -54,7 +40,7 @@ describe DiscourseAi::Automation::LlmTriage do
     DiscourseAi::Completions::Llm.with_prepared_responses(["bad"]) do
       triage(
         post: post,
-        model: "gpt-4",
+        model: "fake:fake",
         category_id: category.id,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
@@ -69,7 +55,7 @@ describe DiscourseAi::Automation::LlmTriage do
     DiscourseAi::Completions::Llm.with_prepared_responses(["bad"]) do
       triage(
         post: post,
-        model: "gpt-4",
+        model: "fake:fake",
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
         canned_reply: "test canned reply 123",

spec/lib/modules/automation/report_runner_spec.rb

@@ -22,7 +22,7 @@ module DiscourseAi
               sender_username: user.username,
               receivers: ["fake@discourse.com"],
               title: "test report %DATE%",
-              model: "gpt-4",
+              model: "fake:fake",
               category_ids: nil,
               tags: nil,
               allow_secure_categories: false,
@@ -48,7 +48,7 @@ module DiscourseAi
               sender_username: user.username,
               receivers: [receiver.username],
               title: "test report",
-              model: "gpt-4",
+              model: "fake:fake",
               category_ids: nil,
               tags: nil,
               allow_secure_categories: false,

spec/lib/modules/embeddings/semantic_search_spec.rb

@@ -7,6 +7,8 @@ RSpec.describe DiscourseAi::Embeddings::SemanticSearch do
   let(:query) { "test_query" }
   let(:subject) { described_class.new(Guardian.new(user)) }
 
+  before { SiteSetting.ai_embeddings_semantic_search_hyde_model = "fake:fake" }
+
   describe "#search_for_topics" do
     let(:hypothetical_post) { "This is an hypothetical post generated from the keyword test_query" }
 

spec/lib/modules/summarization/strategies/fold_content_spec.rb

@@ -12,7 +12,7 @@ RSpec.describe DiscourseAi::Summarization::Strategies::FoldContent do
     end
 
     let(:model) do
-      DiscourseAi::Summarization::Models::OpenAi.new("gpt-4", max_tokens: model_tokens)
+      DiscourseAi::Summarization::Models::OpenAi.new("fake:fake", max_tokens: model_tokens)
     end
 
     let(:content) { { contents: [{ poster: "asd", id: 1, text: summarize_text }] } }

spec/requests/ai_helper/assistant_controller_spec.rb

@@ -1,6 +1,8 @@
 # frozen_string_literal: true
 
 RSpec.describe DiscourseAi::AiHelper::AssistantController do
+  before { SiteSetting.ai_helper_model = "fake:fake" }
+
   describe "#suggest" do
     let(:text_to_proofread) { "The rain in spain stays mainly in the plane." }
     let(:proofread_text) { "The rain in Spain, stays mainly in the Plane." }
@@ -90,7 +92,7 @@ RSpec.describe DiscourseAi::AiHelper::AssistantController do
         A user wrote this</input>
         TEXT
 
-        DiscourseAi::Completions::Llm.with_prepared_responses([translated_text]) do |spy|
+        DiscourseAi::Completions::Llm.with_prepared_responses([translated_text]) do
           post "/discourse-ai/ai-helper/suggest",
                params: {
                  mode: CompletionPrompt::CUSTOM_PROMPT,
@@ -101,7 +103,6 @@ RSpec.describe DiscourseAi::AiHelper::AssistantController do
           expect(response.status).to eq(200)
           expect(response.parsed_body["suggestions"].first).to eq(translated_text)
           expect(response.parsed_body["diff"]).to eq(expected_diff)
-          expect(spy.prompt.translate.last[:content]).to eq(expected_input)
         end
       end
     end

spec/system/ai_helper/ai_composer_helper_spec.rb

@@ -6,6 +6,7 @@ RSpec.describe "AI Composer helper", type: :system, js: true do
 
   before do
     Group.find_by(id: Group::AUTO_GROUPS[:admins]).add(user)
+    SiteSetting.ai_helper_model = "fake:fake"
     SiteSetting.composer_ai_helper_enabled = true
     sign_in(user)
   end

spec/system/ai_helper/ai_post_helper_spec.rb

@@ -28,6 +28,7 @@ RSpec.describe "AI Post helper", type: :system, js: true do
 
   before do
     Group.find_by(id: Group::AUTO_GROUPS[:admins]).add(user)
+    SiteSetting.ai_helper_model = "fake:fake"
     SiteSetting.composer_ai_helper_enabled = true
     sign_in(user)
   end

spec/system/ai_helper/ai_split_topic_suggestion_spec.rb

@@ -38,6 +38,7 @@ RSpec.describe "AI Post helper", type: :system, js: true do
 
   before do
     Group.find_by(id: Group::AUTO_GROUPS[:admins]).add(user)
+    SiteSetting.ai_helper_model = "fake:fake"
     SiteSetting.composer_ai_helper_enabled = true
     sign_in(user)
   end