Enable AI locally ($2554994) · Snippets · GitLab.org / GitLab

@m_gill this script didn't work initially as my GDK directory is called gdk instead of the default gitlab-development-toolkit. Could we change the directory check to check for a file, like GDK_ROOT or gdk.yml?

Everything else worked great! Thank you for putting this together.

If you remove:

if [[ $PWD != *gitlab-development-kit ]]; then
	echo "You must run this script from the gitlab-development-kit directory"
	exit 0
fi

and replace it with:

case "$PWD" in
  *gdk | *gitlab-development-kit) ;;
  *) echo "You must run this script from the gitlab-development-kit directory"
    exit 0
esac

It should work with both gdk and gitlab-development-kit directories 😄

I tried to run the script, but the following line fails if not running in SaaS mode:

Group.first.namespace_settings.update(experiment_features_enabled: true, third_party_ai_features_enabled: true)

The step to enable saas needs to be done before

Interesting catch @eduardobonet. I fixed this in the script 👍

@m_gill I had to make the following adjustments in the script for it to work in my case:

Set -e so execution stops when an error is encountered
Ensure the script commands themselves run with simulated SAAS
Add restarts between ruby scripts where needed for changes to load
Add check that the user has a license set up
Reference gitlab-org group by name instead of Group.first (it wasn't the first one in my db)
Create subscription if it doesn't exist
Use ! methods for updates so errors are more obvious
Use new setup for GCP credentials from !133138 (merged)

@@ -1,5 +1,9 @@
 #!/usr/bin/env bash
 
+set -e
+
+export GITLAB_SIMULATE_SAAS=1
+
 echo "Let's get started setting up your AI environment!"
 echo "This script will configure your GDK to support AI features."
 echo "You will still need to enable any feature-specific feature flags, which you can find here:"
@@ -23,8 +27,8 @@
 echo "What is your OpenAI API key? Leave this blank if you don't need to set OpenAI up right now."
 read -r openai_key
 
-echo "What is the full path to your VertexAI credentials? Leave this blank if you don't need to set VertexAI up right now."
-read -r vertex_credentials_path
+echo "What is your GCP project for Vertex AI API access? Leave this blank if you don't need to set VertexAI up right now."
+read -r vertex_project
 
 echo "Do you want to setup the embeddings database? Y/N"
 read -r setup_embeddings
@@ -57,20 +61,26 @@
        string="export GITLAB_SIMULATE_SAAS=0"
        replace="export GITLAB_SIMULATE_SAAS=1"
        sed -i -e "s/$string/$replace/" $filename
+    gdk restart
 else
        echo "Need to update"
        echo "export GITLAB_SIMULATE_SAAS=1" >> $filename
+    gdk restart
 fi
 
 
 # Setup Ultimate
 
 echo "Being on an Ultimate plan is sometimes a requirement to use AI as a customer."
-echo "The first group in your GDK (usually gitlab-org) will be set to an Ultimate plan."
+echo "Your gitlab-org group will be set to an Ultimate plan."
 
 cd gitlab
 
 cat <<EOF >ensure_ultimate.rb
+if License.count.zero?
+    fail "No license found. Did you follow https://about.gitlab.com/handbook/developer-onboarding/#working-on-gitlab-ee-developer-licenses?"
+end
+
 # Allow use of licensed features
 Gitlab::CurrentSettings.update(check_namespace_plan: true)
 
@@ -80,9 +90,15 @@
        ultimate_plan = Plan.create(name: "ultimate", title: "Ultimate")
 end
 
+group = Group.find_by_full_path('gitlab-org')
 
-Group.first.gitlab_subscription.update(hosted_plan_id: ultimate_plan.id)
+if group.gitlab_subscription
+    group.gitlab_subscription.update!(hosted_plan_id: ultimate_plan.id)
+else
+    GitlabSubscription.create!(namespace: group, hosted_plan_id: ultimate_plan.id)
+end
 
+
 EOF
 
 rails r ensure_ultimate.rb
@@ -92,10 +108,10 @@
 # Setup toggles
 
 echo "Experimental features and third-party AI services are both options that must be enabled to use AI as a customer."
-echo "These settings will be enabled for the first group in your GDK (usually gitlab-org)"
+echo "These settings will be enabled for the gitlab-org group on your GDK"
 
 
-echo "Group.first.namespace_settings.update(experiment_features_enabled: true, third_party_ai_features_enabled: true)" | rails console
+echo "Group.find_by_full_path('gitlab-org').namespace_settings.update!(experiment_features_enabled: true, third_party_ai_features_enabled: true)" | rails console
 cd ../
 
 
@@ -135,14 +151,12 @@
        cd ../
 fi
 
-if [[ $vertex_credentials_path ]]; then
+if [[ $vertex_project ]]; then
        echo "Configuring VertexAI..."
        cd gitlab
 
        cat <<EOF >configure_vertex.rb
-file = File.read('$vertex_credentials_path')
-Gitlab::CurrentSettings.update(vertex_ai_credentials: file )
-Gitlab::CurrentSettings.update(vertex_ai_project: JSON.parse(file)["project_id"])
+Gitlab::CurrentSettings.update(vertex_ai_project: '$vertex_project')
 EOF
 
        rails r configure_vertex.rb
@@ -219,4 +233,3 @@
        cd ../
 
 fi

I replaced the old seeding script with the new rake task and set the file to public:

cat <<EOF >setup_embeddings.rb
ActiveRecord::Base.logger = nil if Rails.env.development?

sha = 'f7bfb0bd48fbb33c620146f5df5a88d54e489127'
number_of_rows = 12739
url = "https://gitlab.com/gitlab-org/enablement-section/tanuki-bot/-/raw/#{sha}/pgvector/tanuki_bot_mvc.json?inline=false"

Dir.mktmpdir do |dir|
  puts "> Loading file size"
  content_length = HTTParty.head(url).headers.fetch('content-length').to_i

  File.open("#{dir}/tanuki_bot_mvc.json", "wb") do |file|
    puts "> Downloading file"
    cursor = 0
    i = 0
    HTTParty.get(url, stream_body: true) do |fragment|
      file.write(fragment)
      cursor += fragment.length
      i += 1
      puts "#{ cursor / 2**20 }MiB (#{((cursor/content_length.to_f) * 100).round(2)}%)" if i % 1000 == 0 || cursor == content_length
    end
  end

  puts "> Reading file"
  Embedding::TanukiBotMvc.transaction do
    Embedding::TanukiBotMvc.delete_all

    i = 0
    File.readlines("#{dir}/tanuki_bot_mvc.json").each do |line|
      attributes = JSON.parse(line)
      Embedding::TanukiBotMvc.create!(attributes)
      i += 1

      count = Embedding::TanukiBotMvc.count
      puts "#{count}/#{number_of_rows} (#{((count/number_of_rows.to_f) * 100).round(2)}%)" if i % 100 == 0 || i == number_of_rows
    end

    # Delete records with no meaningful content
    Embedding::TanukiBotMvc.where.not("content ~ ?", '\w').delete_all
    # Delete records without titles as these likely no longer exist
    Embedding::TanukiBotMvc.where("metadata -> 'title' IS NULL").delete_all

    true
  end

  puts "> Transaction completed"
  puts "Number of records: #{Embedding::TanukiBotMvc.count}"
end
EOF

	rails r setup_embeddings.rb

	rm setup_embeddings.rb

If anyone has issues with postgres when building pgvector/vector eg:

--------------------------------------------------------------------------------
Building pgvector/vector.so version v0.4.1
--------------------------------------------------------------------------------
In file included from src/ivfbuild.c:1:
In file included from /Users/carladrago/.asdf/installs/postgres/14.9/include/server/postgres.h:46:
/Users/carladrago/.asdf/installs/postgres/14.9/include/server/c.h:59:10: fatal error: 'stdio.h' file not found
#include <stdio.h>

uninstalling then reinstalling postgres worked for me:

asdf uninstall postgres 14.9 && asdf install postgres 14.9