dspy 0.30.0 → 0.31.0

data/lib/dspy/lm/adapters/gemini/schema_converter.rb

@@ -1,186 +0,0 @@
-# frozen_string_literal: true
-
-require "sorbet-runtime"
-
-module DSPy
-  class LM
-    module Adapters
-      module Gemini
-        # Converts DSPy signatures to Gemini structured output format
-        class SchemaConverter
-          extend T::Sig
-
-          # Models that support structured outputs (JSON + Schema)
-          # Based on official Google documentation: https://ai.google.dev/gemini-api/docs/models/gemini
-          # Last updated: Oct 2025
-          # Note: Gemini 1.5 series deprecated Oct 2025
-          STRUCTURED_OUTPUT_MODELS = T.let([
-            # Gemini 2.0 series
-            "gemini-2.0-flash",
-            "gemini-2.0-flash-lite",
-            # Gemini 2.5 series (current)
-            "gemini-2.5-pro",
-            "gemini-2.5-flash",
-            "gemini-2.5-flash-lite",
-            "gemini-2.5-flash-image"
-          ].freeze, T::Array[String])
-
-          # Models that do not support structured outputs or are deprecated
-          UNSUPPORTED_MODELS = T.let([
-            # Legacy Gemini 1.0 series
-            "gemini-pro",
-            "gemini-1.0-pro-002",
-            "gemini-1.0-pro",
-            # Deprecated Gemini 1.5 series (removed Oct 2025)
-            "gemini-1.5-pro",
-            "gemini-1.5-pro-preview-0514",
-            "gemini-1.5-pro-preview-0409",
-            "gemini-1.5-flash",
-            "gemini-1.5-flash-8b"
-          ].freeze, T::Array[String])
-
-          sig { params(signature_class: T.class_of(DSPy::Signature)).returns(T::Hash[Symbol, T.untyped]) }
-          def self.to_gemini_format(signature_class)
-            # Get the output JSON schema from the signature class
-            output_schema = signature_class.output_json_schema
-            
-            # Convert to Gemini format (OpenAPI 3.0 Schema subset - not related to OpenAI)
-            convert_dspy_schema_to_gemini(output_schema)
-          end
-
-          sig { params(model: String).returns(T::Boolean) }
-          def self.supports_structured_outputs?(model)
-            # Extract base model name without provider prefix
-            base_model = model.sub(/^gemini\//, "")
-            
-            # Check if it's a supported model or a newer version
-            STRUCTURED_OUTPUT_MODELS.any? { |supported| base_model.start_with?(supported) }
-          end
-
-          sig { params(schema: T::Hash[Symbol, T.untyped]).returns(T::Array[String]) }
-          def self.validate_compatibility(schema)
-            issues = []
-
-            # Check for deeply nested objects (Gemini has depth limits)
-            depth = calculate_depth(schema)
-            if depth > 5
-              issues << "Schema depth (#{depth}) exceeds recommended limit of 5 levels"
-            end
-
-            issues
-          end
-
-          private
-
-          sig { params(dspy_schema: T::Hash[Symbol, T.untyped]).returns(T::Hash[Symbol, T.untyped]) }
-          def self.convert_dspy_schema_to_gemini(dspy_schema)
-            # For Gemini's responseJsonSchema, we need pure JSON Schema format
-            # Remove OpenAPI-specific fields like "$schema"
-            result = {
-              type: "object",
-              properties: {},
-              required: []
-            }
-
-            # Convert properties
-            properties = dspy_schema[:properties] || {}
-            properties.each do |prop_name, prop_schema|
-              result[:properties][prop_name] = convert_property_to_gemini(prop_schema)
-            end
-
-            # Set required fields
-            result[:required] = (dspy_schema[:required] || []).map(&:to_s)
-
-            result
-          end
-
-          sig { params(property_schema: T::Hash[Symbol, T.untyped]).returns(T::Hash[Symbol, T.untyped]) }
-          def self.convert_property_to_gemini(property_schema)
-            # Handle oneOf/anyOf schemas (union types) - Gemini supports these in responseJsonSchema
-            if property_schema[:oneOf]
-              return {
-                oneOf: property_schema[:oneOf].map { |schema| convert_property_to_gemini(schema) },
-                description: property_schema[:description]
-              }.compact
-            end
-            
-            if property_schema[:anyOf]
-              return {
-                anyOf: property_schema[:anyOf].map { |schema| convert_property_to_gemini(schema) },
-                description: property_schema[:description]
-              }.compact
-            end
-            
-            case property_schema[:type]
-            when "string"
-              result = { type: "string" }
-              # Gemini responseJsonSchema doesn't support const, so convert to single-value enum
-              # See: https://ai.google.dev/api/generate-content#FIELDS.response_json_schema
-              if property_schema[:const]
-                result[:enum] = [property_schema[:const]]
-              elsif property_schema[:enum]
-                result[:enum] = property_schema[:enum]
-              end
-              result
-            when "integer"
-              { type: "integer" }
-            when "number"
-              { type: "number" }
-            when "boolean"
-              { type: "boolean" }
-            when "array"
-              {
-                type: "array",
-                items: convert_property_to_gemini(property_schema[:items] || { type: "string" })
-              }
-            when "object"
-              result = { type: "object" }
-              
-              if property_schema[:properties]
-                result[:properties] = {}
-                property_schema[:properties].each do |nested_prop, nested_schema|
-                  result[:properties][nested_prop] = convert_property_to_gemini(nested_schema)
-                end
-                
-                # Set required fields for nested objects
-                if property_schema[:required]
-                  result[:required] = property_schema[:required].map(&:to_s)
-                end
-              end
-              
-              result
-            else
-              # Default to string for unknown types
-              { type: "string" }
-            end
-          end
-
-          sig { params(schema: T::Hash[Symbol, T.untyped], current_depth: Integer).returns(Integer) }
-          def self.calculate_depth(schema, current_depth = 0)
-            return current_depth unless schema.is_a?(Hash)
-
-            max_depth = current_depth
-
-            # Check properties
-            if schema[:properties].is_a?(Hash)
-              schema[:properties].each_value do |prop|
-                if prop.is_a?(Hash)
-                  prop_depth = calculate_depth(prop, current_depth + 1)
-                  max_depth = [max_depth, prop_depth].max
-                end
-              end
-            end
-
-            # Check array items
-            if schema[:items].is_a?(Hash)
-              items_depth = calculate_depth(schema[:items], current_depth + 1)
-              max_depth = [max_depth, items_depth].max
-            end
-
-            max_depth
-          end
-        end
-      end
-    end
-  end
-end

data/lib/dspy/lm/adapters/gemini_adapter.rb

@@ -1,220 +0,0 @@
-# frozen_string_literal: true
-
-require 'gemini-ai'
-require 'json'
-require_relative '../vision_models'
-
-module DSPy
-  class LM
-    class GeminiAdapter < Adapter
-      def initialize(model:, api_key:, structured_outputs: false)
-        super(model: model, api_key: api_key)
-        validate_api_key!(api_key, 'gemini')
-        
-        @structured_outputs_enabled = structured_outputs
-        
-        # Disable streaming for VCR tests since SSE responses don't record properly
-        # But keep streaming enabled for SSEVCR tests (SSE-specific cassettes)
-        @use_streaming = true
-        begin
-          vcr_active = defined?(VCR) && VCR.current_cassette
-          ssevcr_active = defined?(SSEVCR) && SSEVCR.turned_on?
-          
-          # Only disable streaming if regular VCR is active but SSEVCR is not
-          @use_streaming = false if vcr_active && !ssevcr_active
-        rescue
-          # If VCR/SSEVCR is not available or any error occurs, use streaming
-          @use_streaming = true
-        end
-
-        @client = Gemini.new(
-          credentials: {
-            service: 'generative-language-api',
-            api_key: api_key,
-            version: 'v1beta'  # Use beta API version for structured outputs support
-          },
-          options: { 
-            model: model,
-            server_sent_events: @use_streaming
-          }
-        )
-      end
-
-      def chat(messages:, signature: nil, **extra_params, &block)
-        normalized_messages = normalize_messages(messages)
-        
-        # Validate vision support if images are present
-        if contains_images?(normalized_messages)
-          VisionModels.validate_vision_support!('gemini', model)
-          # Convert messages to Gemini format with proper image handling
-          normalized_messages = format_multimodal_messages(normalized_messages)
-        end
-        
-        # Convert DSPy message format to Gemini format
-        gemini_messages = convert_messages_to_gemini_format(normalized_messages)
-        
-        request_params = {
-          contents: gemini_messages
-        }.merge(extra_params)
-
-        begin
-          content = ""
-          final_response_data = nil
-          
-          # Check if we're using streaming or not
-          if @use_streaming
-            # Streaming mode
-            @client.stream_generate_content(request_params) do |chunk|
-              # Handle case where chunk might be a string (from SSE VCR)
-              if chunk.is_a?(String)
-                begin
-                  chunk = JSON.parse(chunk)
-                rescue JSON::ParserError => e
-                  raise AdapterError, "Failed to parse Gemini streaming response: #{e.message}"
-                end
-              end
-              
-              # Extract content from chunks
-              if chunk.dig('candidates', 0, 'content', 'parts')
-                chunk_text = extract_text_from_parts(chunk.dig('candidates', 0, 'content', 'parts'))
-                content += chunk_text
-                
-                # Call block only if provided (for real streaming)
-                block.call(chunk) if block_given?
-              end
-              
-              # Store final response data (usage, metadata) from last chunk
-              if chunk['usageMetadata'] || chunk.dig('candidates', 0, 'finishReason')
-                final_response_data = chunk
-              end
-            end
-          else
-            # Non-streaming mode (for VCR tests)
-            response = @client.generate_content(request_params)
-            
-            # Extract content from single response
-            if response.dig('candidates', 0, 'content', 'parts')
-              content = extract_text_from_parts(response.dig('candidates', 0, 'content', 'parts'))
-            end
-            
-            # Use response as final data
-            final_response_data = response
-          end
-          
-          # Extract usage information from final chunk
-          usage_data = final_response_data&.dig('usageMetadata')
-          usage_struct = usage_data ? UsageFactory.create('gemini', usage_data) : nil
-          
-          # Create metadata from final chunk
-          metadata = {
-            provider: 'gemini',
-            model: model,
-            finish_reason: final_response_data&.dig('candidates', 0, 'finishReason'),
-            safety_ratings: final_response_data&.dig('candidates', 0, 'safetyRatings'),
-            streaming: block_given?
-          }
-          
-          # Create typed metadata
-          typed_metadata = ResponseMetadataFactory.create('gemini', metadata)
-          
-          Response.new(
-            content: content,
-            usage: usage_struct,
-            metadata: typed_metadata
-          )
-        rescue => e
-          handle_gemini_error(e)
-        end
-      end
-
-      private
-
-      # Convert DSPy message format to Gemini format
-      def convert_messages_to_gemini_format(messages)
-        # Gemini expects contents array with role and parts
-        messages.map do |msg|
-          role = case msg[:role]
-                 when 'system'
-                   'user' # Gemini doesn't have explicit system role, merge with user
-                 when 'assistant'
-                   'model'
-                 else
-                   msg[:role]
-                 end
-          
-          if msg[:content].is_a?(Array)
-            # Multimodal content
-            parts = msg[:content].map do |item|
-              case item[:type]
-              when 'text'
-                { text: item[:text] }
-              when 'image'
-                item[:image].to_gemini_format
-              else
-                item
-              end
-            end
-            
-            { role: role, parts: parts }
-          else
-            # Text-only content
-            { role: role, parts: [{ text: msg[:content] }] }
-          end
-        end
-      end
-      
-      # Extract text content from Gemini parts array
-      def extract_text_from_parts(parts)
-        return "" unless parts.is_a?(Array)
-        
-        parts.map { |part| part['text'] }.compact.join
-      end
-      
-      # Format multimodal messages for Gemini
-      def format_multimodal_messages(messages)
-        messages.map do |msg|
-          if msg[:content].is_a?(Array)
-            # Convert multimodal content to Gemini format
-            formatted_content = msg[:content].map do |item|
-              case item[:type]
-              when 'text'
-                { type: 'text', text: item[:text] }
-              when 'image'
-                # Validate image compatibility before formatting
-                item[:image].validate_for_provider!('gemini')
-                item[:image].to_gemini_format
-              else
-                item
-              end
-            end
-            
-            {
-              role: msg[:role],
-              content: formatted_content
-            }
-          else
-            msg
-          end
-        end
-      end
-      
-      # Handle Gemini-specific errors
-      def handle_gemini_error(error)
-        error_msg = error.message.to_s
-        
-        if error_msg.include?('API_KEY') || error_msg.include?('status 400') || error_msg.include?('status 401') || error_msg.include?('status 403')
-          raise AdapterError, "Gemini authentication failed: #{error_msg}. Check your API key."
-        elsif error_msg.include?('RATE_LIMIT') || error_msg.downcase.include?('quota') || error_msg.include?('status 429')
-          raise AdapterError, "Gemini rate limit exceeded: #{error_msg}. Please wait and try again."
-        elsif error_msg.include?('SAFETY') || error_msg.include?('blocked')
-          raise AdapterError, "Gemini content was blocked by safety filters: #{error_msg}"
-        elsif error_msg.include?('image') || error_msg.include?('media')
-          raise AdapterError, "Gemini image processing failed: #{error_msg}. Ensure your image is a valid format and under size limits."
-        else
-          # Generic error handling
-          raise AdapterError, "Gemini adapter error: #{error_msg}"
-        end
-      end
-    end
-  end
-end

data/lib/dspy/lm/adapters/ollama_adapter.rb

@@ -1,73 +0,0 @@
-# frozen_string_literal: true
-
-require 'openai'
-
-module DSPy
-  class LM
-    class OllamaAdapter < OpenAIAdapter
-      DEFAULT_BASE_URL = 'http://localhost:11434/v1'
-      
-      def initialize(model:, api_key: nil, base_url: nil, structured_outputs: true)
-        # Ollama doesn't require API key for local instances
-        # But may need it for remote/protected instances
-        api_key ||= 'ollama' # OpenAI client requires non-empty key
-        base_url ||= DEFAULT_BASE_URL
-        
-        # Store base_url before calling super
-        @base_url = base_url
-        
-        # Don't call parent's initialize, do it manually to control client creation
-        @model = model
-        @api_key = api_key
-        @structured_outputs_enabled = structured_outputs
-        validate_configuration!
-        
-        # Create client with custom base URL
-        @client = OpenAI::Client.new(
-          api_key: @api_key,
-          base_url: @base_url
-        )
-      end
-      
-      def chat(messages:, signature: nil, response_format: nil, &block)
-        # For Ollama, we need to be more lenient with structured outputs
-        # as it may not fully support OpenAI's response_format spec
-        begin
-          super
-        rescue => e
-          # If structured output fails, retry with enhanced prompting
-          if @structured_outputs_enabled && signature && e.message.include?('response_format')
-            DSPy.logger.debug("Ollama structured output failed, falling back to enhanced prompting")
-            @structured_outputs_enabled = false
-            retry
-          else
-            raise
-          end
-        end
-      end
-      
-      private
-      
-      def validate_configuration!
-        super
-        # Additional Ollama-specific validation could go here
-      end
-      
-      def validate_api_key!(api_key, provider)
-        # For Ollama, API key is optional for local instances
-        # Only validate if it looks like a remote URL
-        if @base_url && !@base_url.include?('localhost') && !@base_url.include?('127.0.0.1')
-          super
-        end
-      end
-      
-      
-      # Ollama may have different model support for structured outputs
-      def supports_structured_outputs?
-        # For now, assume all Ollama models support basic JSON mode
-        # but may not support full OpenAI structured output spec
-        true
-      end
-    end
-  end
-end