ruby_llm 1.15.0 → 2.0.0.rc1

data/lib/tasks/models.rake

@@ -1,565 +0,0 @@
-# frozen_string_literal: true
-
-require 'dotenv/load'
-require 'ruby_llm'
-require 'json'
-require 'json-schema'
-require 'fileutils'
-
-desc 'Update models, docs, and aliases'
-task models: ['models:update', 'models:docs', 'models:aliases']
-
-namespace :models do
-  desc 'Update available models from providers (API keys needed)'
-  task :update do
-    puts 'Configuring RubyLLM...'
-    configure_from_env
-    refresh_models
-    display_model_stats
-  end
-
-  desc 'Generate available models documentation'
-  task :docs do
-    FileUtils.mkdir_p('docs/_reference')
-    output = generate_models_markdown
-    File.write('docs/_reference/available-models.md', output)
-    puts 'Generated docs/_reference/available-models.md'
-  end
-
-  desc 'Generate model aliases from registry'
-  task :aliases do
-    generate_aliases
-  end
-end
-
-# Keep aliases:generate for backwards compatibility
-namespace :aliases do
-  task generate: ['models:aliases']
-end
-
-def configure_from_env
-  RubyLLM.configure do |config|
-    config.anthropic_api_key = ENV.fetch('ANTHROPIC_API_KEY', nil)
-    config.azure_api_base = ENV.fetch('AZURE_API_BASE', nil)
-    config.azure_api_key = ENV.fetch('AZURE_API_KEY', nil)
-    config.deepseek_api_key = ENV.fetch('DEEPSEEK_API_KEY', nil)
-    config.gemini_api_key = ENV.fetch('GEMINI_API_KEY', nil)
-    config.mistral_api_key = ENV.fetch('MISTRAL_API_KEY', nil)
-    config.openai_api_key = ENV.fetch('OPENAI_API_KEY', nil)
-    config.openrouter_api_key = ENV.fetch('OPENROUTER_API_KEY', nil)
-    config.perplexity_api_key = ENV.fetch('PERPLEXITY_API_KEY', nil)
-    config.vertexai_location = ENV.fetch('GOOGLE_CLOUD_LOCATION', nil)
-    config.vertexai_project_id = ENV.fetch('GOOGLE_CLOUD_PROJECT', nil)
-    config.xai_api_key = ENV.fetch('XAI_API_KEY', nil)
-    configure_bedrock(config)
-    config.request_timeout = 30
-  end
-end
-
-def configure_bedrock(config)
-  config.bedrock_api_key = ENV.fetch('AWS_ACCESS_KEY_ID', nil)
-  config.bedrock_secret_key = ENV.fetch('AWS_SECRET_ACCESS_KEY', nil)
-  config.bedrock_region = ENV.fetch('AWS_REGION', nil)
-  config.bedrock_session_token = ENV.fetch('AWS_SESSION_TOKEN', nil)
-end
-
-def refresh_models
-  existing_models = RubyLLM::Models.read_from_json
-  initial_count = existing_models.size
-  puts "Refreshing models (#{initial_count} cached)..."
-
-  models = RubyLLM.models.refresh!
-
-  if models.all.empty? && initial_count.zero?
-    puts 'Error: Failed to fetch models.'
-    exit(1)
-  else
-    existing_data = sorted_models_data(existing_models)
-    new_data = sorted_models_data(models.all)
-
-    if new_data == existing_data && initial_count.positive?
-      puts 'Warning: Model list unchanged.'
-    else
-      puts 'Validating models...'
-      validate_models!(models)
-
-      puts "Saving models.json (#{models.all.size} models)"
-      models.save_to_json
-    end
-  end
-
-  @models = models
-end
-
-def sorted_models_data(models)
-  models.map(&:to_h)
-        .sort_by { |model| [model[:provider].to_s, model[:id].to_s] }
-end
-
-def validate_models!(models)
-  schema_path = RubyLLM::Models.schema_file
-  models_data = models.all.map(&:to_h)
-
-  validation_errors = JSON::Validator.fully_validate(schema_path, models_data)
-
-  unless validation_errors.empty?
-    # Save failed models for inspection
-    failed_path = File.expand_path('../ruby_llm/models.failed.json', __dir__)
-    File.write(failed_path, JSON.pretty_generate(models_data))
-
-    puts 'ERROR: Models validation failed:'
-    puts "\nValidation errors:"
-    validation_errors.first(10).each { |error| puts "  - #{error}" }
-    puts "  ... and #{validation_errors.size - 10} more errors" if validation_errors.size > 10
-    puts "-> Failed models saved to: #{failed_path}"
-    exit(1)
-  end
-
-  puts '✓ Models validation passed'
-end
-
-def display_model_stats
-  puts "\nModel count:"
-  provider_counts = @models.all.group_by(&:provider).transform_values(&:count)
-
-  RubyLLM::Provider.providers.each do |sym, provider_class|
-    name = provider_class.name
-    count = provider_counts[sym.to_s] || 0
-    status = status(sym)
-    puts "  #{name}: #{count} models #{status}"
-  end
-
-  puts 'Refresh complete.'
-end
-
-def status(provider_sym)
-  provider_class = RubyLLM::Provider.providers[provider_sym]
-  if provider_class.local?
-    ' (LOCAL - SKIP)'
-  elsif provider_class.configured?(RubyLLM.config)
-    ' (OK)'
-  else
-    ' (NOT CONFIGURED)'
-  end
-end
-
-def generate_models_markdown
-  models = RubyLLM.models.all
-  total_models = models.count
-  provider_count = models.map(&:provider).uniq.count
-  generated_on = Time.now.utc.strftime('%Y-%m-%d')
-
-  <<~MARKDOWN
-    ---
-    layout: default
-    title: Available Models
-    nav_order: 1
-    description: Browse #{total_models} AI models across #{provider_count} providers (not including local providers). Updated #{generated_on}.
-    redirect_from:
-      - /guides/available-models
-    ---
-
-    # {{ page.title }}
-    {: .no_toc }
-
-    {{ page.description }}
-    {: .fs-6 .fw-300 }
-
-    ## Table of contents
-    {: .no_toc .text-delta }
-
-    1. TOC
-    {:toc}
-
-    ---
-
-    _Model information enriched by [models.dev](https://models.dev) and our custom code._
-
-    Can't find a newly released model? Refresh your registry:
-
-    ```ruby
-    # Plain Ruby
-    RubyLLM.models.refresh!
-
-    # Rails
-    Model.refresh!
-    ```
-
-    See [Model Registry: Refreshing the Registry]({% link _advanced/models.md %}#refreshing-the-registry).
-
-    ## Models by Provider
-
-    #{generate_provider_sections}
-
-    ## Models by Capability
-
-    #{generate_capability_sections}
-
-    ## Models by Modality
-
-    #{generate_modality_sections}
-  MARKDOWN
-end
-
-def generate_provider_sections
-  RubyLLM::Provider.providers.filter_map do |provider, provider_class|
-    models = RubyLLM.models.by_provider(provider)
-    next if models.none?
-
-    <<~PROVIDER
-      ### #{provider_class.name} (#{models.count})
-
-      #{models_table(models)}
-    PROVIDER
-  end.join("\n\n")
-end
-
-def generate_capability_sections
-  capabilities = {
-    'Function Calling' => RubyLLM.models.select(&:function_calling?),
-    'Structured Output' => RubyLLM.models.select(&:structured_output?),
-    'Streaming' => RubyLLM.models.select { |m| m.capabilities.include?('streaming') },
-    'Batch Processing' => RubyLLM.models.select { |m| m.capabilities.include?('batch') }
-  }
-
-  capabilities.filter_map do |capability, models|
-    next if models.none?
-
-    <<~CAPABILITY
-      ### #{capability} (#{models.count})
-
-      #{models_table(models)}
-    CAPABILITY
-  end.join("\n\n")
-end
-
-def generate_modality_sections # rubocop:disable Metrics/PerceivedComplexity
-  sections = []
-
-  vision_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('image') }
-  if vision_models.any?
-    sections << <<~SECTION
-      ### Vision Models (#{vision_models.count})
-
-      Models that can process images:
-
-      #{models_table(vision_models)}
-    SECTION
-  end
-
-  audio_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('audio') }
-  if audio_models.any?
-    sections << <<~SECTION
-      ### Audio Input Models (#{audio_models.count})
-
-      Models that can process audio:
-
-      #{models_table(audio_models)}
-    SECTION
-  end
-
-  pdf_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('pdf') }
-  if pdf_models.any?
-    sections << <<~SECTION
-      ### PDF Models (#{pdf_models.count})
-
-      Models that can process PDF documents:
-
-      #{models_table(pdf_models)}
-    SECTION
-  end
-
-  embedding_models = RubyLLM.models.select { |m| (m.modalities.output || []).include?('embeddings') }
-  if embedding_models.any?
-    sections << <<~SECTION
-      ### Embedding Models (#{embedding_models.count})
-
-      Models that generate embeddings:
-
-      #{models_table(embedding_models)}
-    SECTION
-  end
-
-  sections.join("\n\n")
-end
-
-def models_table(models)
-  return '*No models found*' if models.none?
-
-  headers = ['Model', 'Provider', 'I/O', 'Capabilities', 'Context', 'Max Output', 'Standard Pricing (per 1M tokens)']
-  alignment = [':--', ':--', ':--', ':--', '--:', '--:', ':--']
-
-  rows = models.sort_by { |m| [m.provider, m.name] }.map do |model|
-    pricing = standard_pricing_display(model)
-
-    [
-      model.id,
-      model.provider,
-      modalities_display(model),
-      list_display(model.capabilities),
-      model.context_window || '-',
-      model.max_output_tokens || '-',
-      pricing
-    ]
-  end
-
-  table = []
-  table << "| #{headers.join(' | ')} |"
-  table << "| #{alignment.join(' | ')} |"
-
-  rows.each do |row|
-    table << "| #{row.join(' | ')} |"
-  end
-
-  table.join("\n")
-end
-
-def modalities_display(model)
-  input_modalities = list_display(model.modalities.input)
-  output_modalities = list_display(model.modalities.output)
-  "In: #{input_modalities}; Out: #{output_modalities}"
-end
-
-def list_display(values)
-  items = Array(values).compact.map(&:to_s).reject(&:empty?)
-  return '-' if items.empty?
-
-  items.join(', ')
-end
-
-def standard_pricing_display(model)
-  pricing_data = model.pricing.to_h[:text_tokens]&.dig(:standard) || {}
-  parts = [
-    pricing_part(pricing_data, :input_per_million, 'In'),
-    pricing_part(pricing_data, :output_per_million, 'Out'),
-    pricing_part(pricing_data, %i[cache_read_input_per_million cached_input_per_million], 'Cache Read'),
-    pricing_part(pricing_data, %i[cache_write_input_per_million cache_creation_input_per_million], 'Cache Write')
-  ].compact
-
-  return parts.join(', ') if parts.any?
-
-  '-'
-end
-
-def pricing_part(pricing_data, key, label)
-  key = Array(key).find { |candidate| pricing_data[candidate] }
-  return unless key
-
-  "#{label}: $#{format('%.2f', pricing_data[key])}"
-end
-
-def generate_aliases # rubocop:disable Metrics/PerceivedComplexity
-  models = Hash.new { |h, k| h[k] = [] }
-
-  RubyLLM.models.all.each do |model|
-    models[model.provider] << model.id
-  end
-
-  aliases = {}
-
-  # OpenAI models
-  models['openai'].each do |model|
-    openrouter_model = "openai/#{model}"
-    azure_model = models['azure'].include?(model) ? model : nil
-    next unless models['openrouter'].include?(openrouter_model)
-
-    alias_key = model.gsub('-latest', '')
-    aliases[alias_key] = {
-      'openai' => model,
-      'openrouter' => openrouter_model
-    }
-    aliases[alias_key]['azure'] = azure_model if azure_model
-  end
-
-  anthropic_latest = group_anthropic_models_by_base_name(models['anthropic'])
-
-  anthropic_latest.each do |base_name, latest_model|
-    openrouter_variants = [
-      "anthropic/#{base_name}",
-      "anthropic/#{base_name.gsub(/-(\d)/, '.\1')}",
-      "anthropic/#{base_name.gsub(/claude-(\d+)-(\d+)/, 'claude-\1.\2')}",
-      "anthropic/#{base_name.gsub(/(\d+)-(\d+)/, '\1.\2')}"
-    ]
-
-    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
-    bedrock_model = find_best_bedrock_model(latest_model, models['bedrock'])
-
-    next unless openrouter_model || bedrock_model || models['anthropic'].include?(latest_model)
-
-    aliases[base_name] = { 'anthropic' => latest_model }
-    aliases[base_name]['openrouter'] = openrouter_model if openrouter_model
-    aliases[base_name]['bedrock'] = bedrock_model if bedrock_model
-    aliases[base_name]['azure'] = latest_model if models['azure'].include?(latest_model)
-  end
-
-  models['bedrock'].each do |bedrock_model|
-    next unless bedrock_model.start_with?('anthropic.')
-    next unless bedrock_model =~ /anthropic\.(claude-[a-z0-9.-]+)-\d{8}/
-
-    base_name = Regexp.last_match(1)
-    anthropic_name = base_name.tr('.', '-')
-
-    next if aliases[anthropic_name]
-
-    openrouter_variants = [
-      "anthropic/#{anthropic_name}",
-      "anthropic/#{base_name}"
-    ]
-
-    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
-
-    aliases[anthropic_name] = { 'bedrock' => bedrock_model }
-    aliases[anthropic_name]['anthropic'] = anthropic_name if models['anthropic'].include?(anthropic_name)
-    aliases[anthropic_name]['openrouter'] = openrouter_model if openrouter_model
-  end
-
-  # Gemini models (also map to vertexai)
-  models['gemini'].each do |model|
-    openrouter_variants = [
-      "google/#{model}",
-      "google/#{model.gsub('gemini-', 'gemini-').tr('.', '-')}",
-      "google/#{model.gsub('gemini-', 'gemini-')}"
-    ]
-
-    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
-    vertexai_model = models['vertexai'].include?(model) ? model : nil
-
-    next unless openrouter_model || vertexai_model
-
-    alias_key = model.gsub('-latest', '')
-    aliases[alias_key] = { 'gemini' => model }
-    aliases[alias_key]['openrouter'] = openrouter_model if openrouter_model
-    aliases[alias_key]['vertexai'] = vertexai_model if vertexai_model
-  end
-
-  # VertexAI models that aren't in Gemini (e.g. older models like text-bison)
-  models['vertexai'].each do |model|
-    # Skip if already handled above
-    next if models['gemini'].include?(model)
-
-    # Check if OpenRouter has this Google model
-    openrouter_variants = [
-      "google/#{model}",
-      "google/#{model.tr('.', '-')}"
-    ]
-
-    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
-    gemini_model = models['gemini'].include?(model) ? model : nil
-
-    next unless openrouter_model || gemini_model
-
-    alias_key = model.gsub('-latest', '')
-    next if aliases[alias_key] # Skip if already created
-
-    aliases[alias_key] = { 'vertexai' => model }
-    aliases[alias_key]['openrouter'] = openrouter_model if openrouter_model
-    aliases[alias_key]['gemini'] = gemini_model if gemini_model
-  end
-
-  models['deepseek'].each do |model|
-    openrouter_model = "deepseek/#{model}"
-    next unless models['openrouter'].include?(openrouter_model)
-
-    alias_key = model.gsub('-latest', '')
-    aliases[alias_key] = {
-      'deepseek' => model,
-      'openrouter' => openrouter_model
-    }
-  end
-
-  sorted_aliases = aliases.sort.to_h
-  File.write(RubyLLM::Aliases.aliases_file, JSON.pretty_generate(sorted_aliases))
-
-  puts "Generated #{sorted_aliases.size} aliases"
-end
-
-def group_anthropic_models_by_base_name(anthropic_models)
-  grouped = Hash.new { |h, k| h[k] = [] }
-
-  anthropic_models.each do |model|
-    base_name = extract_base_name(model)
-    grouped[base_name] << model
-  end
-
-  latest_models = {}
-  grouped.each do |base_name, model_list|
-    if model_list.size == 1
-      latest_models[base_name] = model_list.first
-    else
-      latest_model = model_list.max_by { |model| extract_date_from_model(model) }
-      latest_models[base_name] = latest_model
-    end
-  end
-
-  latest_models
-end
-
-def extract_base_name(model)
-  if model =~ /^(.+)-(\d{8})$/
-    Regexp.last_match(1)
-  else
-    model
-  end
-end
-
-def extract_date_from_model(model)
-  if model =~ /-(\d{8})$/
-    Regexp.last_match(1)
-  else
-    '00000000'
-  end
-end
-
-def find_best_bedrock_model(anthropic_model, bedrock_models) # rubocop:disable Metrics/PerceivedComplexity
-  base_pattern = case anthropic_model
-                 when 'claude-2.0', 'claude-2'
-                   'claude-v2'
-                 when 'claude-2.1'
-                   'claude-v2:1'
-                 when 'claude-instant-v1', 'claude-instant'
-                   'claude-instant'
-                 else
-                   extract_base_name(anthropic_model)
-                 end
-
-  matching_models = bedrock_models.select do |bedrock_model|
-    model_without_prefix = bedrock_model.sub(/^(?:us\.)?anthropic\./, '')
-    model_without_prefix.start_with?(base_pattern)
-  end
-
-  return nil if matching_models.empty?
-
-  begin
-    model_info = RubyLLM.models.find(anthropic_model)
-    target_context = model_info.context_window
-  rescue StandardError
-    target_context = nil
-  end
-
-  if target_context
-    target_k = target_context / 1000
-
-    with_context = matching_models.select do |m|
-      m.include?(":#{target_k}k") || m.include?(":0:#{target_k}k")
-    end
-
-    return with_context.first if with_context.any?
-  end
-
-  matching_models.min_by do |model|
-    context_priority = if model =~ /:(?:\d+:)?(\d+)k/
-                         -Regexp.last_match(1).to_i
-                       else
-                         0
-                       end
-
-    version_priority = if model =~ /-v(\d+):/
-                         -Regexp.last_match(1).to_i
-                       else
-                         0
-                       end
-
-    has_context_priority = model.include?('k') ? -1 : 0
-    [has_context_priority, context_priority, version_priority]
-  end
-end

data/lib/tasks/release.rake

@@ -1,67 +0,0 @@
-# frozen_string_literal: true
-
-namespace :release do # rubocop:disable Metrics/BlockLength
-  desc 'Prepare for release'
-  task :prepare do
-    Rake::Task['release:refresh_stale_cassettes'].invoke
-    sh 'overcommit --run'
-    Rake::Task['models'].invoke
-  end
-
-  desc 'Remove stale cassettes and re-record them'
-  task :refresh_stale_cassettes do
-    max_age_days = 1
-    cassette_dir = 'spec/fixtures/vcr_cassettes'
-
-    stale_count = 0
-    Dir.glob("#{cassette_dir}/**/*.yml").each do |cassette|
-      age_days = (Time.now - File.mtime(cassette)) / 86_400
-      next unless age_days > max_age_days
-
-      puts "Removing stale cassette: #{File.basename(cassette)} (#{age_days.round(1)} days old)"
-      File.delete(cassette)
-      stale_count += 1
-    end
-
-    if stale_count.positive?
-      puts "\n🗑️  Removed #{stale_count} stale cassettes"
-      puts '🔄 Re-recording cassettes...'
-      run_test_queue_rspec || exit(1)
-      puts '✅ Cassettes refreshed!'
-    else
-      puts '✅ No stale cassettes found'
-    end
-  end
-
-  desc 'Verify cassettes are fresh enough for release'
-  task :verify_cassettes do
-    max_age_days = 1
-    cassette_dir = 'spec/fixtures/vcr_cassettes'
-    stale_cassettes = []
-
-    Dir.glob("#{cassette_dir}/**/*.yml").each do |cassette|
-      age_days = (Time.now - File.mtime(cassette)) / 86_400
-
-      next unless age_days > max_age_days
-
-      stale_cassettes << {
-        file: File.basename(cassette),
-        age: age_days.round(1)
-      }
-    end
-
-    if stale_cassettes.any?
-      puts "\n❌ Found stale cassettes (older than #{max_age_days} days):"
-      stale_files = []
-      stale_cassettes.each do |c|
-        puts "   - #{c[:file]} (#{c[:age]} days old)"
-        stale_files << File.join(cassette_dir, '**', c[:file])
-      end
-
-      puts "\nRun locally: bundle exec rake release:refresh_stale_cassettes"
-      exit 1
-    else
-      puts "✅ All cassettes are fresh (< #{max_age_days} days old)"
-    end
-  end
-end