dspy 0.10.1 → 0.12.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b721c6c31be4c7a18cb488a7646ff356d658618a5a64ff36f50be9f731d9755a
-  data.tar.gz: 2e13ea6bf39e6a89872d604f22c7b197bf50924058824ff2a3483d859f1ba744
+  metadata.gz: d40d9681a87073c6c4f5531025891bf871bee086434ee131e15a4391cba7b893
+  data.tar.gz: 1a1a4ce935e7cdf4ce9c06617f124ed8858d7074a3f6bec4484d6db24dbfa232
 SHA512:
-  metadata.gz: b6966d789e0ebad1e6ee6c65deaef2dec1d67d483908e7c0847df84998b4f167ad346d20de974af7194db890fcccabe38f3b38ab33dde95b430fdcd3b86a6045
-  data.tar.gz: 373e489777b036968ad5152aabc01141212c0409c849d89337075f1b2000de7504b8e2eaccca9755c4e52195b1d05d0667f9e4a8fe54c621cd90ba37b18837b0
+  metadata.gz: 2cdd3ee7e867344b4e45dee7fd0d73dc875ec2c730a2cf75e5f620a19c535f0aac50cc6c88dfb7beb3e075b147e7467ccd4e55a7724f49f3ac111eb080fd88e0
+  data.tar.gz: f5c5ed5dd3b6aa5401c20e31fbbd9638eb1d8c0f3b7fe5adb2e2b69bbb3b051d0ecd277d6e2f3b4d4896fd19f39a89345c9503d97106a3b8e005b1f55a3e5b2b

data/README.md

@@ -1,5 +1,8 @@
 # DSPy.rb
 
+[![Gem Version](https://img.shields.io/gem/v/dspy)](https://rubygems.org/gems/dspy)
+[![Total Downloads](https://img.shields.io/gem/dt/dspy)](https://rubygems.org/gems/dspy)
+
 **Build reliable LLM applications in Ruby using composable, type-safe modules.**
 
 DSPy.rb brings structured LLM programming to Ruby developers. Instead of wrestling with prompt strings and parsing responses, you define typed signatures and compose them into pipelines that just work.
@@ -43,7 +46,7 @@ The result? LLM applications that actually scale and don't break when you sneeze
 
 ## Development Status
 
-DSPy.rb is actively developed and approaching stability at **v0.9.0**. The core framework is production-ready with comprehensive documentation, but I'm battle-testing features through the 0.x series before committing to a stable v1.0 API. 
+DSPy.rb is actively developed and approaching stability at **v0.10.1**. The core framework is production-ready with comprehensive documentation, but I'm battle-testing features through the 0.x series before committing to a stable v1.0 API. 
 
 Real-world usage feedback is invaluable - if you encounter issues or have suggestions, please open a GitHub issue!
 
@@ -174,7 +177,7 @@ DSPy.rb has rapidly evolved from experimental to production-ready:
 
 ## Roadmap - Battle-Testing Toward v1.0
 
-DSPy.rb is currently at **v0.9.0** and approaching stability. I'm focusing on real-world usage and refinement through the 0.10, 0.11, 0.12+ series before committing to a stable v1.0 API.
+DSPy.rb is currently at **v0.10.1** and approaching stability. I'm focusing on real-world usage and refinement through the 0.11, 0.12+ series before committing to a stable v1.0 API.
 
 **Current Focus Areas:**
 - 🚧 **Ollama Support** - Local model integration

data/lib/dspy/errors.rb

@@ -3,6 +3,10 @@
 module DSPy
   class Error < StandardError; end
   
+  class ValidationError < Error; end
+  
+  class DeserializationError < Error; end
+  
   class ConfigurationError < Error
     def self.missing_lm(module_name)
       new(<<~MESSAGE)

data/lib/dspy/example.rb

@@ -77,6 +77,17 @@ module DSPy
       expected_hash
     end
 
+    # Custom equality comparison
+    sig { params(other: T.untyped).returns(T::Boolean) }
+    def ==(other)
+      return false unless other.is_a?(Example)
+      
+      @signature_class == other.signature_class &&
+        input_values == other.input_values &&
+        expected_values == other.expected_values
+    end
+
+
     # Check if prediction matches expected output using struct comparison
     sig { params(prediction: T.untyped).returns(T::Boolean) }
     def matches_prediction?(prediction)
@@ -179,15 +190,6 @@ module DSPy
       examples
     end
 
-    # Equality comparison
-    sig { params(other: T.untyped).returns(T::Boolean) }
-    def ==(other)
-      return false unless other.is_a?(Example)
-      
-      @signature_class == other.signature_class &&
-        input_values == other.input_values &&
-        expected_values == other.expected_values
-    end
 
     # String representation for debugging
     sig { returns(String) }

data/lib/dspy/lm/message_builder.rb

@@ -0,0 +1,28 @@
+# frozen_string_literal: true
+
+module DSPy
+  class LM
+    class MessageBuilder
+      attr_reader :messages
+
+      def initialize
+        @messages = []
+      end
+
+      def system(content)
+        @messages << { role: 'system', content: content.to_s }
+        self
+      end
+
+      def user(content)
+        @messages << { role: 'user', content: content.to_s }
+        self
+      end
+
+      def assistant(content)
+        @messages << { role: 'assistant', content: content.to_s }
+        self
+      end
+    end
+  end
+end

data/lib/dspy/lm.rb

@@ -18,6 +18,9 @@ require_relative 'lm/adapters/anthropic_adapter'
 require_relative 'lm/strategy_selector'
 require_relative 'lm/retry_handler'
 
+# Load message builder
+require_relative 'lm/message_builder'
+
 module DSPy
   class LM
     attr_reader :model_id, :api_key, :model, :provider, :adapter
@@ -39,41 +42,13 @@ module DSPy
       # Build messages from inference module
       messages = build_messages(inference_module, input_values)
       
-      # Calculate input size for monitoring
-      input_text = messages.map { |m| m[:content] }.join(' ')
-      input_size = input_text.length
-      
-      # Use smart consolidation: emit LM events only when not in nested context
-      response = nil
-      token_usage = {}
+      # Execute with instrumentation
+      response = instrument_lm_request(messages, signature_class.name) do
+        chat_with_strategy(messages, signature_class, &block)
+      end
       
+      # Instrument response parsing
       if should_emit_lm_events?
-        # Emit all LM events when not in nested context
-        response = Instrumentation.instrument('dspy.lm.request', {
-          gen_ai_operation_name: 'chat',
-          gen_ai_system: provider,
-          gen_ai_request_model: model,
-          signature_class: signature_class.name,
-          provider: provider,
-          adapter_class: adapter.class.name,
-          input_size: input_size
-        }) do
-          chat_with_strategy(messages, signature_class, &block)
-        end
-        
-        # Extract actual token usage from response (more accurate than estimation)
-        token_usage = Instrumentation::TokenTracker.extract_token_usage(response, provider)
-        
-        # Emit token usage event if available
-        if token_usage.any?
-          Instrumentation.emit('dspy.lm.tokens', token_usage.merge({
-            gen_ai_system: provider,
-            gen_ai_request_model: model,
-            signature_class: signature_class.name
-          }))
-        end
-        
-        # Instrument response parsing
         parsed_result = Instrumentation.instrument('dspy.lm.response.parsed', {
           signature_class: signature_class.name,
           provider: provider,
@@ -82,15 +57,33 @@ module DSPy
           parse_response(response, input_values, signature_class)
         end
       else
-        # Consolidated mode: execute without nested instrumentation
-        response = chat_with_strategy(messages, signature_class, &block)
-        token_usage = Instrumentation::TokenTracker.extract_token_usage(response, provider)
         parsed_result = parse_response(response, input_values, signature_class)
       end
       
       parsed_result
     end
 
+    def raw_chat(messages = nil, &block)
+      # Support both array format and builder DSL
+      if block_given? && messages.nil?
+        # DSL mode - block is for building messages
+        builder = MessageBuilder.new
+        yield builder
+        messages = builder.messages
+        streaming_block = nil
+      else
+        # Array mode - block is for streaming
+        messages ||= []
+        streaming_block = block
+      end
+      
+      # Validate messages format
+      validate_messages!(messages)
+      
+      # Execute with instrumentation
+      execute_raw_chat(messages, &streaming_block)
+    end
+
     private
 
     def chat_with_strategy(messages, signature_class, &block)
@@ -208,5 +201,77 @@ module DSPy
         raise "Failed to parse LLM response as JSON: #{e.message}. Original content length: #{response.content&.length || 0} chars"
       end
     end
+
+    # Common instrumentation method for LM requests
+    def instrument_lm_request(messages, signature_class_name, &execution_block)
+      input_text = messages.map { |m| m[:content] }.join(' ')
+      input_size = input_text.length
+      
+      response = nil
+      
+      if should_emit_lm_events?
+        # Emit dspy.lm.request event
+        response = Instrumentation.instrument('dspy.lm.request', {
+          gen_ai_operation_name: 'chat',
+          gen_ai_system: provider,
+          gen_ai_request_model: model,
+          signature_class: signature_class_name,
+          provider: provider,
+          adapter_class: adapter.class.name,
+          input_size: input_size
+        }, &execution_block)
+        
+        # Extract and emit token usage
+        emit_token_usage(response, signature_class_name)
+      else
+        # Consolidated mode: execute without instrumentation
+        response = execution_block.call
+      end
+      
+      response
+    end
+
+    # Common method to emit token usage events
+    def emit_token_usage(response, signature_class_name)
+      token_usage = Instrumentation::TokenTracker.extract_token_usage(response, provider)
+      
+      if token_usage.any?
+        Instrumentation.emit('dspy.lm.tokens', token_usage.merge({
+          gen_ai_system: provider,
+          gen_ai_request_model: model,
+          signature_class: signature_class_name
+        }))
+      end
+      
+      token_usage
+    end
+
+    def validate_messages!(messages)
+      unless messages.is_a?(Array)
+        raise ArgumentError, "messages must be an array"
+      end
+      
+      valid_roles = %w[system user assistant]
+      
+      messages.each do |message|
+        unless message.is_a?(Hash) && message.key?(:role) && message.key?(:content)
+          raise ArgumentError, "Each message must have :role and :content"
+        end
+        
+        unless valid_roles.include?(message[:role])
+          raise ArgumentError, "Invalid role: #{message[:role]}. Must be one of: #{valid_roles.join(', ')}"
+        end
+      end
+    end
+
+    def execute_raw_chat(messages, &streaming_block)
+      response = instrument_lm_request(messages, 'RawPrompt') do
+        # Direct adapter call, no strategies or JSON parsing
+        adapter.chat(messages: messages, signature: nil, &streaming_block)
+      end
+      
+      # Return raw response content, not parsed JSON
+      response.content
+    end
   end
 end

data/lib/dspy/prediction.rb

@@ -319,6 +319,9 @@ module DSPy
         end
         
         value.each do |k, v|
+          # Skip _type field from being added to the struct (it's not a real field)
+          next if k == :_type || k == "_type"
+          
           prop_info = struct_class.props[k]
           if prop_info
             prop_type = prop_info[:type_object] || prop_info[:type]
@@ -343,7 +346,27 @@ module DSPy
           value
         end
       when T::Types::Union
-        # For unions without discriminator, try each type
+        # Check if value has a _type field for automatic type detection
+        type_name = value[:_type] || value["_type"]
+        
+        if type_name
+          # Use _type field to determine which struct to instantiate
+          type.types.each do |t|
+            next if t == T::Utils.coerce(NilClass)
+            
+            if t.is_a?(T::Types::Simple) && t.raw_type < T::Struct
+              struct_name = t.raw_type.name.split('::').last
+              if struct_name == type_name
+                return convert_to_struct(value, t)
+              end
+            end
+          end
+          
+          # If no matching type found, raise an error
+          raise DSPy::DeserializationError, "Unknown type: #{type_name}. Expected one of: #{type.types.map { |t| t.is_a?(T::Types::Simple) && t.raw_type < T::Struct ? t.raw_type.name.split('::').last : nil }.compact.join(', ')}"
+        end
+        
+        # Fallback to trying each type if no _type field
         type.types.each do |t|
           next if t == T::Utils.coerce(NilClass)
           
@@ -398,7 +421,27 @@ module DSPy
 
     sig { params(hash: T::Hash[Symbol, T.untyped], union_type: T::Types::Union).returns(T.untyped) }
     def convert_hash_to_union_struct(hash, union_type)
-      # Try to match the hash structure to one of the union types
+      # First check if hash has a _type field for automatic type detection
+      type_name = hash[:_type] || hash["_type"]
+      
+      if type_name
+        # Use _type field to determine which struct to instantiate
+        union_type.types.each do |type|
+          next if type == T::Utils.coerce(NilClass)
+          
+          if type.is_a?(T::Types::Simple) && type.raw_type < T::Struct
+            struct_name = type.raw_type.name.split('::').last
+            if struct_name == type_name
+              return convert_to_struct(hash, type)
+            end
+          end
+        end
+        
+        # If no matching type found, raise an error
+        raise DSPy::DeserializationError, "Unknown type: #{type_name}. Expected one of: #{union_type.types.map { |t| t.is_a?(T::Types::Simple) && t.raw_type < T::Struct ? t.raw_type.name.split('::').last : nil }.compact.join(', ')}"
+      end
+      
+      # Fallback: Try to match the hash structure to one of the union types
       union_type.types.each do |type|
         next if type == T::Utils.coerce(NilClass)
         
@@ -412,6 +455,9 @@ module DSPy
               # Need to convert nested values too
               converted_hash = {}
               hash.each do |k, v|
+                # Skip _type field
+                next if k == :_type || k == "_type"
+                
                 prop_info = struct_class.props[k]
                 if prop_info
                   prop_type = prop_info[:type_object] || prop_info[:type]

data/lib/dspy/re_act.rb

@@ -256,7 +256,7 @@ module DSPy
       }) do
         # Generate thought and action
         thought_obj = @thought_generator.forward(
-          input_context: input_struct.serialize.to_json,
+          input_context: DSPy::TypeSerializer.serialize(input_struct).to_json,
           history: history,
           available_tools: available_tools_desc
         )
@@ -385,7 +385,7 @@ module DSPy
       return { should_finish: false } if observation.include?("Unknown action")
 
       observation_result = @observation_processor.forward(
-        input_context: input_struct.serialize.to_json,
+        input_context: DSPy::TypeSerializer.serialize(input_struct).to_json,
         history: history,
         observation: observation
       )
@@ -402,7 +402,7 @@ module DSPy
     sig { params(input_struct: T.untyped, history: T::Array[HistoryEntry], available_tools_desc: T::Array[T::Hash[String, T.untyped]], observation_result: T.untyped, iteration: Integer).returns(String) }
     def generate_forced_final_answer(input_struct, history, available_tools_desc, observation_result, iteration)
       final_thought = @thought_generator.forward(
-        input_context: input_struct.serialize.to_json,
+        input_context: DSPy::TypeSerializer.serialize(input_struct).to_json,
         history: history,
         available_tools: available_tools_desc
       )

data/lib/dspy/signature.rb

@@ -292,6 +292,19 @@ module DSPy
         properties = {}
         required = []
 
+        # Check if struct already has a _type field
+        if struct_class.props.key?(:_type)
+          raise DSPy::ValidationError, "_type field conflict: #{struct_class.name} already has a _type field defined. " \
+                                       "DSPy uses _type for automatic type detection in union types."
+        end
+
+        # Add automatic _type field for type detection
+        properties[:_type] = {
+          type: "string",
+          const: struct_class.name.split('::').last  # Use the simple class name
+        }
+        required << "_type"
+
         struct_class.props.each do |prop_name, prop_info|
           prop_type = prop_info[:type_object] || prop_info[:type]
           properties[prop_name] = type_to_json_schema(prop_type)

data/lib/dspy/type_serializer.rb

@@ -0,0 +1,55 @@
+# frozen_string_literal: true
+
+require "sorbet-runtime"
+
+module DSPy
+  class TypeSerializer
+    extend T::Sig
+
+    # Serialize a value, injecting _type fields for T::Struct instances
+    sig { params(value: T.untyped).returns(T.untyped) }
+    def self.serialize(value)
+      case value
+      when T::Struct
+        serialize_struct(value)
+      when Array
+        value.map { |item| serialize(item) }
+      when Hash
+        value.transform_values { |v| serialize(v) }
+      else
+        value
+      end
+    end
+
+    private
+
+    sig { params(struct: T::Struct).returns(T::Hash[String, T.untyped]) }
+    def self.serialize_struct(struct)
+      # Handle anonymous structs that don't have a name
+      class_name = struct.class.name
+      type_name = if class_name.nil? || class_name.empty?
+        # For anonymous structs, use a generic identifier
+        "AnonymousStruct"
+      else
+        class_name.split('::').last
+      end
+      
+      result = {
+        "_type" => type_name
+      }
+
+      # Get all props and serialize their values
+      struct.class.props.each do |prop_name, prop_info|
+        prop_value = struct.send(prop_name)
+        
+        # Skip nil values for optional fields
+        next if prop_value.nil? && prop_info[:fully_optional]
+        
+        # Recursively serialize nested values
+        result[prop_name.to_s] = serialize(prop_value)
+      end
+
+      result
+    end
+  end
+end

data/lib/dspy/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module DSPy
-  VERSION = "0.10.1"
+  VERSION = "0.12.0"
 end

data/lib/dspy.rb

@@ -6,6 +6,7 @@ require 'securerandom'
 
 require_relative 'dspy/version'
 require_relative 'dspy/errors'
+require_relative 'dspy/type_serializer'
 
 module DSPy
   extend Dry::Configurable

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: dspy
 version: !ruby/object:Gem::Version
-  version: 0.10.1
+  version: 0.12.0
 platform: ruby
 authors:
 - Vicente Reig Rincón de Arellano
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2025-07-20 00:00:00.000000000 Z
+date: 2025-07-24 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: dry-configurable
@@ -177,6 +177,7 @@ files:
 - lib/dspy/lm/adapters/openai_adapter.rb
 - lib/dspy/lm/cache_manager.rb
 - lib/dspy/lm/errors.rb
+- lib/dspy/lm/message_builder.rb
 - lib/dspy/lm/response.rb
 - lib/dspy/lm/retry_handler.rb
 - lib/dspy/lm/strategies/anthropic_extraction_strategy.rb
@@ -223,6 +224,7 @@ files:
 - lib/dspy/tools/memory_toolset.rb
 - lib/dspy/tools/text_processing_toolset.rb
 - lib/dspy/tools/toolset.rb
+- lib/dspy/type_serializer.rb
 - lib/dspy/version.rb
 homepage: https://github.com/vicentereig/dspy.rb
 licenses: