ruby_llm 1.16.0 → 2.0.0.rc2

data/lib/ruby_llm/providers/gemini/chat.rb

@@ -1,542 +0,0 @@
-# frozen_string_literal: true
-
-require 'set'
-require 'rubygems/version'
-
-module RubyLLM
-  module Providers
-    class Gemini
-      # Chat methods for the Gemini API implementation
-      module Chat
-        module_function
-
-        def completion_url
-          "models/#{@model}:generateContent"
-        end
-
-        # rubocop:disable Metrics/ParameterLists,Lint/UnusedMethodArgument
-        def render_payload(messages, tools:, temperature:, model:, stream: false, schema: nil,
-                           thinking: nil, tool_prefs: nil)
-          tool_prefs ||= {}
-          @model = model.id
-          payload = {
-            contents: format_messages(messages),
-            generationConfig: {}
-          }
-
-          payload[:generationConfig][:temperature] = temperature unless temperature.nil?
-
-          payload[:generationConfig].merge!(structured_output_config(schema, model)) if schema
-          payload[:generationConfig][:thinkingConfig] = build_thinking_config(model, thinking) if thinking&.enabled?
-
-          if tools.any?
-            payload[:tools] = format_tools(tools)
-            # Gemini doesn't support controlling parallel tool calls
-            payload[:toolConfig] = build_tool_config(tool_prefs[:choice]) unless tool_prefs[:choice].nil?
-          end
-
-          payload
-        end
-        # rubocop:enable Metrics/ParameterLists,Lint/UnusedMethodArgument
-
-        def build_thinking_config(_model, thinking)
-          config = { includeThoughts: true }
-
-          config[:thinkingLevel] = resolve_effort_level(thinking) if thinking&.effort
-          config[:thinkingBudget] = resolve_budget(thinking) if thinking&.budget
-
-          config
-        end
-
-        def resolve_effort_level(thinking)
-          thinking.respond_to?(:effort) ? thinking.effort : thinking
-        end
-
-        def resolve_budget(thinking)
-          budget = thinking.respond_to?(:budget) ? thinking.budget : thinking
-          budget.is_a?(Integer) ? budget : nil
-        end
-
-        private
-
-        def format_messages(messages)
-          formatter = MessageFormatter.new(
-            messages,
-            format_role: method(:format_role),
-            format_parts: method(:format_parts),
-            format_tool_result: method(:format_tool_result)
-          )
-          formatter.format
-        end
-
-        def format_role(role)
-          case role
-          when :assistant then 'model'
-          when :system, :tool then 'user'
-          else role.to_s
-          end
-        end
-
-        def format_parts(msg)
-          if msg.tool_call?
-            format_tool_call(msg)
-          elsif msg.tool_result?
-            format_tool_result(msg)
-          else
-            format_message_parts(msg)
-          end
-        end
-
-        def format_message_parts(msg)
-          parts = []
-
-          parts << build_thought_part(msg.thinking) if msg.role == :assistant && msg.thinking
-
-          content_parts = Media.format_content(msg.content)
-          parts.concat(content_parts.is_a?(Array) ? content_parts : [content_parts])
-          parts
-        end
-
-        def build_thought_part(thinking)
-          part = { thought: true }
-          part[:text] = thinking.text if thinking.text
-          part[:thoughtSignature] = thinking.signature if thinking.signature
-          part
-        end
-
-        def parse_completion_response(response)
-          data = response.body
-          parts = data.dig('candidates', 0, 'content', 'parts') || []
-          tool_calls = extract_tool_calls(data)
-
-          Message.new(
-            role: :assistant,
-            content: extract_text_parts(parts) || parse_content(data),
-            thinking: Thinking.build(
-              text: extract_thought_parts(parts),
-              signature: extract_thought_signature(parts)
-            ),
-            tool_calls: tool_calls,
-            input_tokens: input_tokens(data),
-            output_tokens: calculate_output_tokens(data),
-            cached_tokens: data.dig('usageMetadata', 'cachedContentTokenCount'),
-            thinking_tokens: data.dig('usageMetadata', 'thoughtsTokenCount'),
-            model_id: data['modelVersion'] || response.env.url.path.split('/')[3].split(':')[0],
-            raw: response
-          )
-        end
-
-        def input_tokens(data)
-          prompt_tokens = data.dig('usageMetadata', 'promptTokenCount')
-          return unless prompt_tokens
-
-          [prompt_tokens.to_i - data.dig('usageMetadata', 'cachedContentTokenCount').to_i, 0].max
-        end
-
-        def convert_schema_to_gemini(schema)
-          return nil unless schema
-
-          # Extract inner schema if wrapper format (e.g., from RubyLLM::Schema.to_json_schema)
-          schema = schema[:schema] || schema
-
-          GeminiSchema.new(schema).to_h
-        end
-
-        def parse_content(data)
-          candidate = data.dig('candidates', 0)
-          return '' unless candidate
-
-          return '' if function_call?(candidate)
-
-          parts = candidate.dig('content', 'parts')
-          return '' unless parts&.any?
-
-          non_thought_parts = parts.reject { |part| part['thought'] }
-          return '' unless non_thought_parts.any?
-
-          build_response_content(non_thought_parts)
-        end
-
-        def extract_text_parts(parts)
-          text_parts = parts.reject { |p| p['thought'] }
-          content = text_parts.filter_map { |p| p['text'] }.join
-          content.empty? ? nil : content
-        end
-
-        def extract_thought_parts(parts)
-          thought_parts = parts.select { |p| p['thought'] }
-          thoughts = thought_parts.filter_map { |p| p['text'] }.join
-          thoughts.empty? ? nil : thoughts
-        end
-
-        def extract_thought_signature(parts)
-          parts.each do |part|
-            signature = part['thoughtSignature'] ||
-                        part['thought_signature'] ||
-                        part.dig('functionCall', 'thoughtSignature') ||
-                        part.dig('functionCall', 'thought_signature')
-            return signature if signature
-          end
-
-          nil
-        end
-
-        def function_call?(candidate)
-          parts = candidate.dig('content', 'parts')
-          parts&.any? { |p| p['functionCall'] }
-        end
-
-        def calculate_output_tokens(data)
-          candidates = data.dig('usageMetadata', 'candidatesTokenCount') || 0
-          thoughts = data.dig('usageMetadata', 'thoughtsTokenCount') || 0
-          candidates + thoughts
-        end
-
-        def response_json_schema_supported?(model)
-          version = gemini_version(model)
-          version && version >= Gem::Version.new('2.5')
-        end
-
-        def build_json_schema(schema)
-          normalized = RubyLLM::Utils.deep_dup(schema[:schema])
-          normalized.delete(:strict)
-          normalized.delete('strict')
-          RubyLLM::Utils.deep_stringify_keys(normalized)
-        end
-
-        def gemini_version(model)
-          return nil unless model
-
-          candidates = [
-            safe_string(model.id),
-            safe_string(model.respond_to?(:family) ? model.family : nil),
-            safe_string(model_metadata_value(model, :version)),
-            safe_string(model_metadata_value(model, 'version')),
-            safe_string(model_metadata_value(model, :description))
-          ].compact
-
-          candidates.each do |candidate|
-            version = extract_version(candidate)
-            return version if version
-          end
-
-          nil
-        end
-
-        def model_metadata_value(model, key)
-          return unless model.respond_to?(:metadata)
-
-          metadata = model.metadata
-          return unless metadata.is_a?(Hash)
-
-          metadata[key] || metadata[key.to_s]
-        end
-
-        def safe_string(value)
-          value&.to_s
-        end
-
-        def extract_version(text)
-          return nil unless text
-
-          match = text.match(/(\d+\.\d+|\d+)/)
-          return nil unless match
-
-          Gem::Version.new(match[1])
-        rescue ArgumentError
-          nil
-        end
-
-        def structured_output_config(schema, model)
-          {
-            responseMimeType: 'application/json'
-          }.tap do |config|
-            if response_json_schema_supported?(model)
-              config[:responseJsonSchema] = build_json_schema(schema)
-            else
-              config[:responseSchema] = convert_schema_to_gemini(schema)
-            end
-          end
-        end
-
-        # formats a message
-        class MessageFormatter
-          def initialize(messages, format_role:, format_parts:, format_tool_result:)
-            @messages = messages
-            @index = 0
-            @tool_call_names = {}
-            @format_role = format_role
-            @format_parts = format_parts
-            @format_tool_result = format_tool_result
-          end
-
-          def format
-            formatted = []
-
-            while current_message
-              if tool_message?(current_message)
-                tool_parts, next_index = collect_tool_parts
-                formatted << build_tool_response(tool_parts)
-                @index = next_index
-              else
-                remember_tool_calls if current_message.tool_call?
-                formatted << build_standard_message(current_message)
-                @index += 1
-              end
-            end
-
-            formatted
-          end
-
-          private
-
-          def current_message
-            @messages[@index]
-          end
-
-          def tool_message?(message)
-            message&.role == :tool
-          end
-
-          def collect_tool_parts
-            parts = []
-            index = @index
-
-            while tool_message?(@messages[index])
-              tool_message = @messages[index]
-              tool_name = @tool_call_names.delete(tool_message.tool_call_id)
-              parts.concat(format_tool_result(tool_message, tool_name))
-              index += 1
-            end
-
-            [parts, index]
-          end
-
-          def build_tool_response(parts)
-            { role: 'user', parts: parts }
-          end
-
-          def remember_tool_calls
-            current_message.tool_calls.each do |tool_call_id, tool_call|
-              @tool_call_names[tool_call_id] = tool_call.name
-            end
-          end
-
-          def build_standard_message(message)
-            {
-              role: @format_role.call(message.role),
-              parts: @format_parts.call(message)
-            }
-          end
-
-          def format_tool_result(message, tool_name)
-            @format_tool_result.call(message, tool_name)
-          end
-        end
-
-        # converts json schema to gemini
-        class GeminiSchema
-          def initialize(schema)
-            @raw_schema = RubyLLM::Utils.deep_dup(schema)
-            @definitions = {}
-          end
-
-          def to_h
-            return nil unless @raw_schema
-
-            symbolized = symbolize_and_extract_definitions(@raw_schema)
-            convert(symbolized, Set.new)
-          end
-
-          private
-
-          attr_reader :definitions
-
-          def symbolize_and_extract_definitions(value)
-            case value
-            when Hash
-              value.each_with_object({}) do |(key, val), hash|
-                key_sym = begin
-                  key.to_sym
-                rescue StandardError
-                  key
-                end
-
-                if definition_key?(key_sym)
-                  merge_definitions(val)
-                else
-                  hash[key_sym] = symbolize_and_extract_definitions(val)
-                end
-              end
-            when Array
-              value.map { |item| symbolize_and_extract_definitions(item) }
-            else
-              value
-            end
-          end
-
-          def definition_key?(key)
-            %i[$defs definitions].include?(key)
-          end
-
-          def merge_definitions(raw_defs)
-            return unless raw_defs
-
-            symbolized = symbolize_and_extract_definitions(raw_defs)
-            @definitions = if definitions.empty?
-                             symbolized
-                           else
-                             RubyLLM::Utils.deep_merge(definitions, symbolized)
-                           end
-          end
-
-          def convert(schema, visited_refs)
-            return default_string_schema unless schema.is_a?(Hash)
-
-            schema = strip_unsupported_keys(schema)
-
-            if schema[:$ref]
-              resolved = resolve_reference(schema, visited_refs)
-              return resolved if resolved
-            end
-
-            schema = normalize_any_of(schema)
-
-            result = case schema[:type].to_s
-                     when 'object'
-                       build_object(schema, visited_refs)
-                     when 'array'
-                       build_array(schema, visited_refs)
-                     when 'number'
-                       build_scalar('NUMBER', schema, %i[format minimum maximum enum nullable multipleOf])
-                     when 'integer'
-                       build_scalar('INTEGER', schema, %i[format minimum maximum enum nullable multipleOf])
-                     when 'boolean'
-                       build_scalar('BOOLEAN', schema, %i[nullable])
-                     else
-                       build_scalar('STRING', schema, %i[enum format nullable])
-                     end
-
-            apply_description(result, schema)
-            result
-          end
-
-          def strip_unsupported_keys(schema)
-            schema.dup.tap do |copy|
-              copy.delete(:strict)
-              copy.delete(:additionalProperties)
-            end
-          end
-
-          def resolve_reference(schema, visited_refs)
-            ref = schema[:$ref]
-            return unless ref
-            return if visited_refs.include?(ref)
-
-            referenced = lookup_definition(ref)
-            return unless referenced
-
-            overrides = schema.except(:$ref)
-            visited_refs.add(ref)
-            merged = RubyLLM::Utils.deep_merge(referenced, overrides)
-            convert(merged, visited_refs)
-          ensure
-            visited_refs.delete(ref)
-          end
-
-          def lookup_definition(ref) # rubocop:disable Metrics/PerceivedComplexity
-            segments = ref.to_s.split('/').reject(&:empty?)
-            return nil if segments.empty?
-
-            segments.shift if segments.first == '#'
-            segments.shift if %w[$defs definitions].include?(segments.first)
-
-            current = definitions
-
-            segments.each do |segment|
-              break current = nil unless current.is_a?(Hash)
-
-              key = begin
-                segment.to_sym
-              rescue StandardError
-                segment
-              end
-              current = current[key]
-            end
-
-            current ? RubyLLM::Utils.deep_dup(current) : nil
-          end
-
-          def normalize_any_of(schema)
-            any_of = schema[:anyOf]
-            return schema unless any_of
-
-            options = Array(any_of).map { |option| RubyLLM::Utils.deep_symbolize_keys(option) }
-            nullables, non_null = options.partition { |option| schema_type(option) == 'null' }
-
-            base = RubyLLM::Utils.deep_symbolize_keys(non_null.first || { type: 'string' })
-            base[:nullable] = true if nullables.any?
-
-            without_any_of = schema.each_with_object({}) do |(key, value), result|
-              result[key] = value unless key == :anyOf
-            end
-
-            without_any_of.merge(base)
-          end
-
-          def schema_type(option)
-            (option[:type] || option['type']).to_s.downcase
-          end
-
-          def build_object(schema, visited_refs)
-            properties = schema.fetch(:properties, {}).transform_values do |child|
-              convert(child, visited_refs)
-            end
-
-            {
-              type: 'OBJECT',
-              properties: properties
-            }.tap do |object|
-              required = Array(schema[:required]).map(&:to_s).uniq
-              object[:required] = required if required.any?
-              object[:propertyOrdering] = schema[:propertyOrdering] if schema[:propertyOrdering]
-              copy_attribute(object, schema, :nullable)
-            end
-          end
-
-          def build_array(schema, visited_refs)
-            items_schema = schema[:items] ? convert(schema[:items], visited_refs) : default_string_schema
-
-            {
-              type: 'ARRAY',
-              items: items_schema
-            }.tap do |array|
-              copy_attribute(array, schema, :minItems)
-              copy_attribute(array, schema, :maxItems)
-              copy_attribute(array, schema, :nullable)
-            end
-          end
-
-          def build_scalar(type, schema, allowed_keys)
-            { type: type }.tap do |result|
-              allowed_keys.each { |key| copy_attribute(result, schema, key) }
-            end
-          end
-
-          def apply_description(target, schema)
-            description = schema[:description]
-            target[:description] = description if description
-          end
-
-          def copy_attribute(target, source, key)
-            target[key] = source[key] if source.key?(key)
-          end
-
-          def default_string_schema
-            { type: 'STRING' }
-          end
-        end
-      end
-    end
-  end
-end

data/lib/ruby_llm/providers/gemini/embeddings.rb

@@ -1,37 +0,0 @@
-# frozen_string_literal: true
-
-module RubyLLM
-  module Providers
-    class Gemini
-      # Embeddings methods for the Gemini API integration
-      module Embeddings
-        module_function
-
-        def embedding_url(model:)
-          "models/#{model}:batchEmbedContents"
-        end
-
-        def render_embedding_payload(text, model:, dimensions:)
-          { requests: [text].flatten.map { |t| single_embedding_payload(t, model:, dimensions:) } }
-        end
-
-        def parse_embedding_response(response, model:, text:)
-          vectors = response.body['embeddings']&.map { |e| e['values'] }
-          vectors = vectors.first if vectors&.length == 1 && !text.is_a?(Array)
-
-          Embedding.new(vectors:, model:, input_tokens: 0)
-        end
-
-        private
-
-        def single_embedding_payload(text, model:, dimensions:)
-          {
-            model: "models/#{model}",
-            content: { parts: [{ text: text.to_s }] },
-            outputDimensionality: dimensions
-          }.compact
-        end
-      end
-    end
-  end
-end

data/lib/ruby_llm/providers/gemini/images.rb

@@ -1,47 +0,0 @@
-# frozen_string_literal: true
-
-module RubyLLM
-  module Providers
-    class Gemini
-      # Image generation methods for the Gemini API implementation
-      module Images
-        def images_url(with: nil, mask: nil) # rubocop:disable Lint/UnusedMethodArgument
-          "models/#{@model}:predict"
-        end
-
-        def render_image_payload(prompt, model:, size:, with: nil, mask: nil, params: {}) # rubocop:disable Lint/UnusedMethodArgument,Metrics/ParameterLists
-          RubyLLM.logger.debug { "Ignoring size #{size}. Gemini does not support image size customization." }
-          @model = model
-          {
-            instances: [
-              {
-                prompt: prompt
-              }
-            ],
-            parameters: {
-              sampleCount: 1
-            }
-          }
-        end
-
-        def parse_image_response(response, model:)
-          data = response.body
-          image_data = data['predictions']&.first
-
-          unless image_data&.key?('bytesBase64Encoded')
-            raise Error, 'Unexpected response format from Gemini image generation API'
-          end
-
-          mime_type = image_data['mimeType'] || 'image/png'
-          base64_data = image_data['bytesBase64Encoded']
-
-          Image.new(
-            data: base64_data,
-            mime_type: mime_type,
-            model_id: model
-          )
-        end
-      end
-    end
-  end
-end

data/lib/ruby_llm/providers/gemini/models.rb

@@ -1,38 +0,0 @@
-# frozen_string_literal: true
-
-module RubyLLM
-  module Providers
-    class Gemini
-      # Models methods for the Gemini API integration
-      module Models
-        module_function
-
-        def models_url
-          'models'
-        end
-
-        def parse_list_models_response(response, slug, capabilities)
-          Array(response.body['models']).map do |model_data|
-            model_id = model_data['name'].gsub('models/', '')
-
-            Model::Info.new(
-              id: model_id,
-              name: model_data['displayName'] || model_id,
-              provider: slug,
-              created_at: nil,
-              context_window: model_data['inputTokenLimit'] || capabilities.context_window_for(model_id),
-              max_output_tokens: model_data['outputTokenLimit'] || capabilities.max_tokens_for(model_id),
-              capabilities: capabilities.critical_capabilities_for(model_id),
-              pricing: capabilities.pricing_for(model_id),
-              metadata: {
-                version: model_data['version'],
-                description: model_data['description'],
-                supported_generation_methods: model_data['supportedGenerationMethods']
-              }
-            )
-          end
-        end
-      end
-    end
-  end
-end