langchainrb 0.6.10 → 0.6.12

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ba1fb0e3fbc05e4279fe3a698ad8fb1a25e02788991a8e6a7b27b411771096f3
-  data.tar.gz: 309cda1c8c7a4982b22c6ad2f82c20fb12ca3bdfdc3e8c0ebeaa9687a7f71ce0
+  metadata.gz: d4567e8a572ad802d06af2571fd5be51a0376a7507f74474eb0598198b5cc29a
+  data.tar.gz: 1032626da89febc17cbf76db2f37a17fbfe53398291cd8fa6e3726d621b9f429
 SHA512:
-  metadata.gz: 3457cbad7efbc5504f4cb3b684e3837984be5d486c1ee21a718508d606dd63ccc2223dffc55f3cc9c52f5fd0a533b364407bbb4c3208515e2bab8ca2af9ea60a
-  data.tar.gz: 1ff3cded239c286ee87d7a2f0a1cfa45c734b185e35b6e6fbc63fd2951f7c7397e8562fa145a16f545ef07b2e983f40715010e9c96653b5f2bf128325a5a7577
+  metadata.gz: 1119378271d50091473d3999c16c78479da6fe7609e1583d3a2f964e7edc1f4802ee98646e4a22d784cd600cdbfdff2da9313ca492f77d2f2743a6ae6082f9bd
+  data.tar.gz: c85e6f0b76bb982546531cc9a2403c4bb9de60b9270078739c1591a563ab6e5404bb94177cef7281e9cbfdfb2938386f3bbdfdd182d202d3bedabef2dfd1383c

data/CHANGELOG.md

@@ -1,5 +1,9 @@
 ## [Unreleased]
 
+## [0.6.12] - 2023-08-13
+
+## [0.6.11] - 2023-08-08
+
 ## [0.6.10] - 2023-08-01
 - 🗣️ LLMs
   - Introducing Anthropic support

data/README.md

@@ -6,7 +6,7 @@
 
 :warning: UNDER ACTIVE AND RAPID DEVELOPMENT (MAY BE BUGGY AND UNTESTED)
 
-![Tests status](https://github.com/andreibondarev/langchainrb/actions/workflows/ci.yml/badge.svg)
+![Tests status](https://github.com/andreibondarev/langchainrb/actions/workflows/ci.yml/badge.svg?branch=main)
 [![Gem Version](https://badge.fury.io/rb/langchainrb.svg)](https://badge.fury.io/rb/langchainrb)
 [![Docs](http://img.shields.io/badge/yard-docs-blue.svg)](http://rubydoc.info/gems/langchainrb)
 [![License](https://img.shields.io/badge/license-MIT-green.svg)](https://github.com/andreibondarev/langchainrb/blob/main/LICENSE.txt)
@@ -161,11 +161,8 @@ qdrant:
 
 ```ruby
 client.llm.functions = functions
-client.llm.complete_response = true
 ```
 
-`complete_response` will return the entire choices data from the gpt response
-
 #### Cohere
 Add `gem "cohere-ruby", "~> 0.9.3"` to your Gemfile.
 
@@ -209,6 +206,10 @@ Add `gem "anthropic", "~> 0.1.0"` to your Gemfile.
 anthropic = Langchain::LLM::Anthropic.new(api_key: ENV["ANTHROPIC_API_KEY"])
 ```
 
+```ruby
+anthropic.complete(prompt: "What is the meaning of life?")
+```
+
 ### Using Prompts 📋
 
 #### Prompt Templates
@@ -418,9 +419,6 @@ agent = Langchain::Agent::ReActAgent.new(
   llm: openai,
   tools: [search_tool, calculator]
 )
-
-agent.tools
-# => ["google_search", "calculator"]
 ```
 ```ruby
 agent.run(question: "How many full soccer fields would be needed to cover the distance between NYC and DC in a straight line?")

data/lib/langchain/agent/base.rb

@@ -7,6 +7,7 @@ module Langchain::Agent
   #
   # Available:
   # - {Langchain::Agent::ReActAgent}
+  # - {Langchain::Agent::SQLQueryAgent}
   #
   # @abstract
   class Base

data/lib/langchain/agent/{react_agent/react_agent.rb → react_agent.rb}

@@ -7,12 +7,13 @@ module Langchain::Agent
   #
   #     agent = Langchain::Agent::ReActAgent.new(
   #       llm: llm,
-  #       tools: ["google_search", "calculator", "wikipedia"]
+  #       tools: [
+  #         Langchain::Tool::GoogleSearch.new(api_key: "YOUR_API_KEY"),
+  #         Langchain::Tool::Calculator.new,
+  #         Langchain::Tool::Wikipedia.new
+  #       ]
   #     )
   #
-  #     agent.tools
-  #     # => ["google_search", "calculator", "wikipedia"]
-  #
   #     agent.run(question: "How many full soccer fields would be needed to cover the distance between NYC and DC in a straight line?")
   #     #=> "Approximately 2,945 soccer fields would be needed to cover the distance between NYC and DC in a straight line."
   class ReActAgent < Base
@@ -21,7 +22,7 @@ module Langchain::Agent
     # Initializes the Agent
     #
     # @param llm [Object] The LLM client to use
-    # @param tools [Array] The tools to use
+    # @param tools [Array<Tool>] The tools to use
     # @param max_iterations [Integer] The maximum number of iterations to run
     # @return [ReActAgent] The Agent::ReActAgent instance
     def initialize(llm:, tools: [], max_iterations: 10)
@@ -35,8 +36,8 @@ module Langchain::Agent
 
     # Validate tools when they're re-assigned
     #
-    # @param value [Array] The tools to use
-    # @return [Array] The tools that will be used
+    # @param value [Array<Tool>] The tools to use
+    # @return [Array<Tool>] The tools that will be used
     def tools=(value)
       Langchain::Tool::Base.validate_tools!(tools: value)
       @tools = value
@@ -70,7 +71,7 @@ module Langchain::Agent
           action_input = response.match(/Action Input: "?(.*)"?/)&.send(:[], -1)
 
           # Find the Tool and call `execute`` with action_input as the input
-          tool = tools.find { |tool| tool.tool_name == action.strip }
+          tool = tools.find { |tool| tool.name == action.strip }
           Langchain.logger.info("Invoking \"#{tool.class}\" Tool with \"#{action_input}\"", for: self.class)
 
           # Call `execute` with action_input as the input
@@ -99,15 +100,15 @@ module Langchain::Agent
     # @param tools [Array] Tools to use
     # @return [String] Prompt
     def create_prompt(question:, tools:)
-      tool_list = tools.map(&:tool_name)
+      tool_list = tools.map(&:name)
 
       prompt_template.format(
         date: Date.today.strftime("%B %d, %Y"),
         question: question,
         tool_names: "[#{tool_list.join(", ")}]",
         tools: tools.map do |tool|
-          tool_name = tool.tool_name
-          tool_description = tool.tool_description
+          tool_name = tool.name
+          tool_description = tool.description
           "#{tool_name}: #{tool_description}"
         end.join("\n")
       )

data/lib/langchain/ai_message.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+module Langchain
+  class AIMessage < Message
+    def type
+      "ai"
+    end
+  end
+end

data/lib/langchain/conversation.rb

@@ -39,45 +39,45 @@ module Langchain
 
     def set_functions(functions)
       @llm.functions = functions
-      @llm.complete_response = true
     end
 
     # Set the context of the conversation. Usually used to set the model's persona.
     # @param message [String] The context of the conversation
     def set_context(message)
-      @memory.set_context message
+      @memory.set_context SystemMessage.new(message)
     end
 
     # Add examples to the conversation. Used to give the model a sense of the conversation.
-    # @param examples [Array<Hash>] The examples to add to the conversation
+    # @param examples [Array<AIMessage|HumanMessage>] The examples to add to the conversation
     def add_examples(examples)
       @memory.add_examples examples
     end
 
     # Message the model with a prompt and return the response.
     # @param message [String] The prompt to message the model with
-    # @return [String] The response from the model
+    # @return [AIMessage] The response from the model
     def message(message)
-      @memory.append_user_message(message)
-      response = llm_response(message)
-      @memory.append_ai_message(response)
-      response
+      human_message = HumanMessage.new(message)
+      @memory.append_message(human_message)
+      ai_message = llm_response(human_message)
+      @memory.append_message(ai_message)
+      ai_message
     end
 
     # Messages from conversation memory
-    # @return [Array<Hash>] The messages from the conversation memory
+    # @return [Array<AIMessage|HumanMessage>] The messages from the conversation memory
     def messages
       @memory.messages
     end
 
     # Context from conversation memory
-    # @return [String] Context from conversation memory
+    # @return [SystemMessage] Context from conversation memory
     def context
       @memory.context
     end
 
     # Examples from conversation memory
-    # @return [Array<Hash>] Examples from the conversation memory
+    # @return [Array<AIMessage|HumanMessage>] Examples from the conversation memory
     def examples
       @memory.examples
     end

data/lib/langchain/conversation_memory.rb

@@ -25,12 +25,8 @@ module Langchain
       @examples.concat examples
     end
 
-    def append_ai_message(message)
-      @messages << {role: "ai", content: message}
-    end
-
-    def append_user_message(message)
-      @messages << {role: "user", content: message}
+    def append_message(message)
+      @messages.append(message)
     end
 
     def reduce_messages(exception)
@@ -47,7 +43,7 @@ module Langchain
     def context
       return if @context.nil? && @summary.nil?
 
-      [@context, @summary].compact.join("\n")
+      SystemMessage.new([@context, @summary].compact.join("\n"))
     end
 
     private

data/lib/langchain/human_message.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+module Langchain
+  class HumanMessage < Message
+    def type
+      "human"
+    end
+  end
+end

data/lib/langchain/llm/anthropic.rb

@@ -13,7 +13,8 @@ module Langchain::LLM
   class Anthropic < Base
     DEFAULTS = {
       temperature: 0.0,
-      completion_model_name: "claude-2"
+      completion_model_name: "claude-2",
+      max_tokens_to_sample: 256
     }.freeze
 
     # TODO: Implement token length validator for Anthropic
@@ -49,7 +50,7 @@ module Langchain::LLM
     private
 
     def compose_parameters(model, params)
-      default_params = {model: model, temperature: @defaults[:temperature]}
+      default_params = {model: model}.merge(@defaults.except(:completion_model_name))
 
       default_params.merge(params)
     end

data/lib/langchain/llm/cohere.rb

@@ -70,7 +70,8 @@ module Langchain::LLM
 
     # Cohere does not have a dedicated chat endpoint, so instead we call `complete()`
     def chat(...)
-      complete(...)
+      response_text = complete(...)
+      Langchain::AIMessage.new(response_text)
     end
 
     # Generate a summary in English for a given text

data/lib/langchain/llm/google_palm.rb

@@ -19,6 +19,9 @@ module Langchain::LLM
       embeddings_model_name: "embedding-gecko-001"
     }.freeze
     LENGTH_VALIDATOR = Langchain::Utils::TokenLength::GooglePalmValidator
+    ROLE_MAPPING = {
+      "human" => "user"
+    }
 
     def initialize(api_key:, default_options: {})
       depends_on "google_palm_api"
@@ -72,10 +75,12 @@ module Langchain::LLM
     #
     # Generate a chat completion for a given prompt
     #
-    # @param prompt [String] The prompt to generate a chat completion for
-    # @param messages [Array] The messages that have been sent in the conversation
-    # @param params extra parameters passed to GooglePalmAPI::Client#generate_chat_message
-    # @return [String] The chat completion
+    # @param prompt [HumanMessage] The prompt to generate a chat completion for
+    # @param messages [Array<AIMessage|HumanMessage>] The messages that have been sent in the conversation
+    # @param context [SystemMessage] An initial context to provide as a system message, ie "You are RubyGPT, a helpful chat bot for helping people learn Ruby"
+    # @param examples [Array<AIMessage|HumanMessage>] Examples of messages to provide to the model. Useful for Few-Shot Prompting
+    # @param options [Hash] extra parameters passed to GooglePalmAPI::Client#generate_chat_message
+    # @return [AIMessage] The chat completion
     #
     def chat(prompt: "", messages: [], context: "", examples: [], **options)
       raise ArgumentError.new(":prompt or :messages argument is expected") if prompt.empty? && messages.empty?
@@ -83,7 +88,7 @@ module Langchain::LLM
       default_params = {
         temperature: @defaults[:temperature],
         model: @defaults[:chat_completion_model_name],
-        context: context,
+        context: context.to_s,
         messages: compose_chat_messages(prompt: prompt, messages: messages),
         examples: compose_examples(examples)
       }
@@ -104,7 +109,7 @@ module Langchain::LLM
       response = client.generate_chat_message(**default_params)
       raise "GooglePalm API returned an error: #{response}" if response.dig("error")
 
-      response.dig("candidates", 0, "content")
+      Langchain::AIMessage.new(response.dig("candidates", 0, "content"))
     end
 
     #
@@ -146,8 +151,8 @@ module Langchain::LLM
     def compose_examples(examples)
       examples.each_slice(2).map do |example|
         {
-          input: {content: example.first[:content]},
-          output: {content: example.last[:content]}
+          input: {content: example.first.content},
+          output: {content: example.last.content}
         }
       end
     end
@@ -155,8 +160,8 @@ module Langchain::LLM
     def transform_messages(messages)
       messages.map do |message|
         {
-          author: message[:role] || message["role"],
-          content: message[:content] || message["content"]
+          author: ROLE_MAPPING.fetch(message.type, message.type),
+          content: message.content
         }
       end
     end

data/lib/langchain/llm/llama_cpp.rb

@@ -33,8 +33,8 @@ module Langchain::LLM
       @seed = seed
     end
 
-    # @params text [String] The text to embed
-    # @params n_threads [Integer] The number of CPU threads to use
+    # @param text [String] The text to embed
+    # @param n_threads [Integer] The number of CPU threads to use
     # @return [Array] The embedding
     def embed(text:, n_threads: nil)
       # contexts are kinda stateful when it comes to embeddings, so allocate one each time
@@ -49,9 +49,9 @@ module Langchain::LLM
       context.embeddings
     end
 
-    # @params prompt [String] The prompt to complete
-    # @params n_predict [Integer] The number of tokens to predict
-    # @params n_threads [Integer] The number of CPU threads to use
+    # @param prompt [String] The prompt to complete
+    # @param n_predict [Integer] The number of tokens to predict
+    # @param n_threads [Integer] The number of CPU threads to use
     # @return [String] The completed prompt
     def complete(prompt:, n_predict: 128, n_threads: nil)
       n_threads ||= self.n_threads

data/lib/langchain/llm/openai.rb

@@ -18,8 +18,12 @@ module Langchain::LLM
       dimension: 1536
     }.freeze
     LENGTH_VALIDATOR = Langchain::Utils::TokenLength::OpenAIValidator
+    ROLE_MAPPING = {
+      "ai" => "assistant",
+      "human" => "user"
+    }
 
-    attr_accessor :functions, :complete_response
+    attr_accessor :functions
 
     def initialize(api_key:, llm_options: {}, default_options: {})
       depends_on "ruby-openai"
@@ -98,19 +102,13 @@ module Langchain::LLM
     #         },
     #       ]
     #
-    # @param prompt [String] The prompt to generate a chat completion for
-    # @param messages [Array<Hash>] The messages that have been sent in the conversation
-    #   Each message should be a Hash with the following keys:
-    #   - :content [String] The content of the message
-    #   - :role [String] The role of the sender (system, user, assistant, or function)
-    # @param context [String] An initial context to provide as a system message, ie "You are RubyGPT, a helpful chat bot for helping people learn Ruby"
-    # @param examples [Array<Hash>] Examples of messages to provide to the model. Useful for Few-Shot Prompting
-    #   Each message should be a Hash with the following keys:
-    #   - :content [String] The content of the message
-    #   - :role [String] The role of the sender (system, user, assistant, or function)
-    # @param options <Hash> extra parameters passed to OpenAI::Client#chat
-    # @yield [String] Stream responses back one String at a time
-    # @return [String] The chat completion
+    # @param prompt [HumanMessage] The prompt to generate a chat completion for
+    # @param messages [Array<AIMessage|HumanMessage>] The messages that have been sent in the conversation
+    # @param context [SystemMessage] An initial context to provide as a system message, ie "You are RubyGPT, a helpful chat bot for helping people learn Ruby"
+    # @param examples [Array<AIMessage|HumanMessage>] Examples of messages to provide to the model. Useful for Few-Shot Prompting
+    # @param options [Hash] extra parameters passed to OpenAI::Client#chat
+    # @yield [AIMessage] Stream responses back one String at a time
+    # @return [AIMessage] The chat completion
     #
     def chat(prompt: "", messages: [], context: "", examples: [], **options)
       raise ArgumentError.new(":prompt or :messages argument is expected") if prompt.empty? && messages.empty?
@@ -126,16 +124,20 @@ module Langchain::LLM
 
       if (streaming = block_given?)
         parameters[:stream] = proc do |chunk, _bytesize|
-          yield chunk if complete_response
-          yield chunk.dig("choices", 0, "delta", "content") if !complete_response
+          delta = chunk.dig("choices", 0, "delta")
+          content = delta["content"]
+          additional_kwargs = {function_call: delta["function_call"]}.compact
+          yield Langchain::AIMessage.new(content, additional_kwargs)
         end
       end
 
       response = client.chat(parameters: parameters)
       raise Langchain::LLM::ApiError.new "Chat completion failed: #{response.dig("error", "message")}" if !response.empty? && response.dig("error")
       unless streaming
-        return response.dig("choices", 0, "message", "content") if !complete_response
-        return response if complete_response
+        message = response.dig("choices", 0, "message")
+        content = message["content"]
+        additional_kwargs = {function_call: message["function_call"]}.compact
+        Langchain::AIMessage.new(content.to_s, additional_kwargs)
       end
     end
 
@@ -171,9 +173,9 @@ module Langchain::LLM
 
       history.concat transform_messages(messages) unless messages.empty?
 
-      unless context.nil? || context.empty?
+      unless context.nil? || context.to_s.empty?
         history.reject! { |message| message[:role] == "system" }
-        history.prepend({role: "system", content: context})
+        history.prepend({role: "system", content: context.content})
       end
 
       unless prompt.empty?
@@ -189,12 +191,9 @@ module Langchain::LLM
 
     def transform_messages(messages)
       messages.map do |message|
-        role = message[:role] || message["role"]
-        content = message[:content] || message["content"]
-
         {
-          content: content,
-          role: (role == "ai") ? "assistant" : role
+          role: ROLE_MAPPING.fetch(message.type, message.type),
+          content: message.content
         }
       end
     end

data/lib/langchain/llm/replicate.rb

@@ -84,7 +84,8 @@ module Langchain::LLM
 
     # Cohere does not have a dedicated chat endpoint, so instead we call `complete()`
     def chat(...)
-      complete(...)
+      response_text = complete(...)
+      Langchain::AIMessage.new(response_text)
     end
 
     #

data/lib/langchain/loader.rb

@@ -89,9 +89,9 @@ module Langchain
     end
 
     def load_from_path
-      raise FileNotFound unless File.exist?(@path)
+      return File.open(@path) if File.exist?(@path)
 
-      File.open(@path)
+      raise FileNotFound, "File #{@path} does not exist"
     end
 
     def load_from_directory(&block)

data/lib/langchain/message.rb

@@ -0,0 +1,35 @@
+# frozen_string_literal: true
+
+module Langchain
+  class Message
+    attr_reader :content, :additional_kwargs
+
+    def initialize(content, additional_kwargs = nil)
+      @content = content
+      @additional_kwargs = additional_kwargs
+    end
+
+    def type
+      raise NotImplementedError
+    end
+
+    def to_s
+      content
+    end
+
+    def ==(other)
+      to_json == other.to_json
+    end
+
+    def to_json(options = {})
+      hash = {
+        type: type,
+        content: content
+      }
+
+      hash[:additional_kwargs] = additional_kwargs unless additional_kwargs.nil? || additional_kwargs.empty?
+
+      hash.to_json
+    end
+  end
+end

data/lib/langchain/output_parsers/base.rb

@@ -9,7 +9,8 @@ module Langchain::OutputParsers
     # Parse the output of an LLM call.
     #
     # @param text - LLM output to parse.
-    # @returns Parsed output.
+    #
+    # @return [Object] Parsed output.
     #
     def parse(text:)
       raise NotImplementedError
@@ -18,9 +19,9 @@ module Langchain::OutputParsers
     #
     # Return a string describing the format of the output.
     #
-    # @returns Format instructions.
-    # @param options - Options for formatting instructions.
-    # @example
+    # @return [String] Format instructions.
+    #
+    # @example returns the format instructions
     # ```json
     # {
     #  "foo": "bar"

data/lib/langchain/output_parsers/{fix.rb → output_fixing_parser.rb}

@@ -65,7 +65,9 @@ module Langchain::OutputParsers
     #
     # Creates a new instance of the class using the given JSON::Schema.
     #
-    # @param schema [JSON::Schema] The JSON::Schema to use
+    # @param llm [Langchain::LLM] The LLM used in the fixing process
+    # @param parser [Langchain::OutputParsers] The parser originally used which resulted in parsing error
+    # @param prompt [Langchain::Prompt::PromptTemplate]
     #
     # @return [Object] A new instance of the class
     #