langchainrb 0.6.16 → 0.6.18

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 36e0bec4ad6abfd9077c9e7f2d6166ba99acb7dc3859749ee6facfb9409e6379
-  data.tar.gz: 6bd8d3de4f1d31b718381fcef1c21a8b417b2bd8483d7fdc2610cfda3b60a50e
+  metadata.gz: 437c6387ded139ed1a513414bfb7242cdbadf1ba6526c7a89346aa2fa9490fc2
+  data.tar.gz: dd6f437a4bbc4807a16631dd790f66c9de4e9456011b2c4f84302fe3fab1377b
 SHA512:
-  metadata.gz: ed7be8f193d44075f701622fd991127ab32580293fb6d1ab7ccc096eeff8704312ad34cdb7a4cfd09cf8879116ede17a5b017fe15851b9ee78cb159b7e8d8b59
-  data.tar.gz: f70d7a3707ed7fce123c2f9158c338cda3aa38a46abf5598f7d05c6ccd63d5a16a37ba10ff0a7a0a4cd17c0c2aeb2f07a07842a41f16322c48c7c9bae522dda4
+  metadata.gz: 24748539de50dfa816fdb71173ef00a6b04f9737f32926fca919865a49b9812dd9f1fdb286c361c98e33cc994f67e8988ab688bfdf6bf3020d954eb0c791177c
+  data.tar.gz: 283b10460187cada7485e08a19c89e7485925ab2f73a5ad51b06a72e8fd9ee1600ddac9d000f13c0c1af13f6defece9fdcc272489d0df803f94da96fe1c76cfd

data/CHANGELOG.md

@@ -1,5 +1,16 @@
 ## [Unreleased]
 
+## [0.6.18] - 2023-10-16
+- Introduce `Langchain::LLM::Response`` object
+- Introduce `Langchain::Chunk` object
+- Add the ask() method to the Langchain::ActiveRecord::Hooks
+
+## [0.6.17] - 2023-10-10
+- Bump weaviate and chroma-db deps
+- `Langchain::Chunker::Semantic` chunker
+- Re-structure Conversations class
+- Bug fixes
+
 ## [0.6.16] - 2023-10-02
 - HyDE-style similarity search
 - `Langchain::Chunker::Sentence` chunker

data/README.md

@@ -59,7 +59,7 @@ client = Langchain::Vectorsearch::Weaviate.new(
 )
 
 # You can instantiate any other supported vector search database:
-client = Langchain::Vectorsearch::Chroma.new(...) # `gem "chroma-db", "~> 0.3.0"`
+client = Langchain::Vectorsearch::Chroma.new(...) # `gem "chroma-db", "~> 0.6.0"`
 client = Langchain::Vectorsearch::Hnswlib.new(...) # `gem "hnswlib", "~> 0.8.1"`
 client = Langchain::Vectorsearch::Milvus.new(...) # `gem "milvus", "~> 0.9.2"`
 client = Langchain::Vectorsearch::Pinecone.new(...) # `gem "pinecone", "~> 0.1.6"`
@@ -128,6 +128,21 @@ class Product < ActiveRecord::Base
 end
 ```
 
+### Exposed ActiveRecord methods
+```ruby
+# Retrieve similar products based on the query string passed in
+Product.similarity_search(
+    query:,
+    k:       # number of results to be retrieved
+)
+```
+```ruby
+# Q&A-style querying based on the question passed in
+Product.ask(
+    question:
+)
+```
+
 Additional info [here](https://github.com/andreibondarev/langchainrb/blob/main/lib/langchain/active_record/hooks.rb#L10-L38).
 
 ### Using Standalone LLMs 🗣️

data/lib/langchain/active_record/hooks.rb

@@ -92,6 +92,20 @@ module Langchain
           ids = records.map { |record| record.dig("id") || record.dig("__id") }
           where(id: ids)
         end
+
+        # Ask a question and return the answer
+        #
+        # @param question [String] The question to ask
+        # @param k [Integer] The number of results to have in context
+        # @yield [String] Stream responses back one String at a time
+        # @return [String] The answer to the question
+        def ask(question:, k: 4, &block)
+          class_variable_get(:@@provider).ask(
+            question: question,
+            k: k,
+            &block
+          )
+        end
       end
     end
   end

data/lib/langchain/agent/react_agent.rb

@@ -58,7 +58,7 @@ module Langchain::Agent
       max_iterations.times do
         Langchain.logger.info("Sending the prompt to the #{llm.class} LLM", for: self.class)
 
-        response = llm.complete(prompt: prompt, stop_sequences: ["Observation:"])
+        response = llm.complete(prompt: prompt, stop_sequences: ["Observation:"]).completion
 
         # Append the response to the prompt
         prompt += response

data/lib/langchain/agent/sql_query_agent.rb

@@ -27,7 +27,7 @@ module Langchain::Agent
 
       # Get the SQL string to execute
       Langchain.logger.info("Passing the inital prompt to the #{llm.class} LLM", for: self.class)
-      sql_string = llm.complete(prompt: prompt)
+      sql_string = llm.complete(prompt: prompt).completion
 
       # Execute the SQL string and collect the results
       Langchain.logger.info("Passing the SQL to the Database: #{sql_string}", for: self.class)
@@ -36,7 +36,7 @@ module Langchain::Agent
       # Pass the results and get the LLM to synthesize the answer to the question
       Langchain.logger.info("Passing the synthesize prompt to the #{llm.class} LLM with results: #{results}", for: self.class)
       prompt2 = create_prompt_for_answer(question: question, sql_query: sql_string, results: results)
-      llm.complete(prompt: prompt2)
+      llm.complete(prompt: prompt2).completion
     end
 
     private

data/lib/langchain/chunk.rb

@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+
+module Langchain
+  class Chunk
+    # The chunking process is the process of splitting a document into smaller chunks and creating instances of Langchain::Chunk
+
+    attr_reader :text
+
+    # Initialize a new chunk
+    # @param [String] text
+    # @return [Langchain::Chunk]
+    def initialize(text:)
+      @text = text
+    end
+  end
+end

data/lib/langchain/chunker/base.rb

@@ -8,8 +8,15 @@ module Langchain
     #
     # == Available chunkers
     #
+    # - {Langchain::Chunker::RecursiveText}
     # - {Langchain::Chunker::Text}
+    # - {Langchain::Chunker::Semantic}
+    # - {Langchain::Chunker::Sentence}
     class Base
+      # @return [Array<Langchain::Chunk>]
+      def chunks
+        raise NotImplementedError
+      end
     end
   end
 end

data/lib/langchain/chunker/prompts/semantic_prompt_template.yml

@@ -0,0 +1,8 @@
+_type: prompt
+input_variables:
+  - text
+template: | 
+  Please split the following text by topics.
+  Output only the paragraphs delimited by "---":
+
+  {text}

data/lib/langchain/chunker/recursive_text.rb

@@ -24,14 +24,17 @@ module Langchain
         @separators = separators
       end
 
-      # @return [Array<String>]
+      # @return [Array<Langchain::Chunk>]
       def chunks
         splitter = Baran::RecursiveCharacterTextSplitter.new(
           chunk_size: chunk_size,
           chunk_overlap: chunk_overlap,
           separators: separators
         )
-        splitter.chunks(text)
+
+        splitter.chunks(text).map do |chunk|
+          Langchain::Chunk.new(text: chunk[:text])
+        end
       end
     end
   end

data/lib/langchain/chunker/semantic.rb

@@ -0,0 +1,52 @@
+# frozen_string_literal: true
+
+module Langchain
+  module Chunker
+    #
+    # LLM-powered semantic chunker.
+    # Semantic chunking is a technique of splitting texts by their semantic meaning, e.g.: themes, topics, and ideas.
+    # We use an LLM to accomplish this. The Anthropic LLM is highly recommended for this task as it has the longest context window (100k tokens).
+    #
+    # Usage:
+    #     Langchain::Chunker::Semantic.new(
+    #       text,
+    #       llm: Langchain::LLM::Anthropic.new(api_key: ENV["ANTHROPIC_API_KEY"])
+    #     ).chunks
+    #
+    class Semantic < Base
+      attr_reader :text, :llm, :prompt_template
+      # @param [Langchain::LLM::Base] Langchain::LLM::* instance
+      # @param [Langchain::Prompt::PromptTemplate] Optional custom prompt template
+      def initialize(text, llm:, prompt_template: nil)
+        @text = text
+        @llm = llm
+        @prompt_template = prompt_template || default_prompt_template
+      end
+
+      # @return [Array<Langchain::Chunk>]
+      def chunks
+        prompt = prompt_template.format(text: text)
+
+        # Replace static 50k limit with dynamic limit based on text length (max_tokens_to_sample)
+        completion = llm.complete(prompt: prompt, max_tokens_to_sample: 50000)
+        completion
+          .gsub("Here are the paragraphs split by topic:\n\n", "")
+          .split("---")
+          .map(&:strip)
+          .reject(&:empty?)
+          .map do |chunk|
+            Langchain::Chunk.new(text: chunk)
+          end
+      end
+
+      private
+
+      # @return [Langchain::Prompt::PromptTemplate] Default prompt template for semantic chunking
+      def default_prompt_template
+        Langchain::Prompt.load_from_path(
+          file_path: Langchain.root.join("langchain/chunker/prompts/semantic_prompt_template.yml")
+        )
+      end
+    end
+  end
+end

data/lib/langchain/chunker/sentence.rb

@@ -19,10 +19,12 @@ module Langchain
         @text = text
       end
 
-      # @return [Array<String>]
+      # @return [Array<Langchain::Chunk>]
       def chunks
         ps = PragmaticSegmenter::Segmenter.new(text: text)
-        ps.segment
+        ps.segment.map do |chunk|
+          Langchain::Chunk.new(text: chunk)
+        end
       end
     end
   end

data/lib/langchain/chunker/text.rb

@@ -24,14 +24,17 @@ module Langchain
         @separator = separator
       end
 
-      # @return [Array<String>]
+      # @return [Array<Langchain::Chunk>]
       def chunks
         splitter = Baran::CharacterTextSplitter.new(
           chunk_size: chunk_size,
           chunk_overlap: chunk_overlap,
           separator: separator
         )
-        splitter.chunks(text)
+
+        splitter.chunks(text).map do |chunk|
+          Langchain::Chunk.new(text: chunk[:text])
+        end
       end
     end
   end

data/lib/langchain/{ai_message.rb → conversation/context.rb}

@@ -1,9 +1,8 @@
 # frozen_string_literal: true
 
 module Langchain
-  class AIMessage < Message
-    def type
-      "ai"
+  class Conversation
+    class Context < Message
     end
   end
 end

data/lib/langchain/conversation/memory.rb

@@ -0,0 +1,86 @@
+# frozen_string_literal: true
+
+module Langchain
+  class Conversation
+    class Memory
+      attr_reader :examples, :messages
+
+      # The least number of tokens we want to be under the limit by
+      TOKEN_LEEWAY = 20
+
+      def initialize(llm:, messages: [], **options)
+        @llm = llm
+        @context = nil
+        @summary = nil
+        @examples = []
+        @messages = messages
+        @strategy = options.delete(:strategy) || :truncate
+        @options = options
+      end
+
+      def set_context(message)
+        @context = message
+      end
+
+      def add_examples(examples)
+        @examples.concat examples
+      end
+
+      def append_message(message)
+        @messages.append(message)
+      end
+
+      def reduce_messages(exception)
+        case @strategy
+        when :truncate
+          truncate_messages(exception)
+        when :summarize
+          summarize_messages
+        else
+          raise "Unknown strategy: #{@options[:strategy]}"
+        end
+      end
+
+      def context
+        return if @context.nil? && @summary.nil?
+
+        Context.new([@context, @summary].compact.join("\n"))
+      end
+
+      private
+
+      def truncate_messages(exception)
+        raise exception if @messages.size == 1
+
+        token_overflow = exception.token_overflow
+
+        @messages = @messages.drop_while do |message|
+          proceed = token_overflow > -TOKEN_LEEWAY
+          token_overflow -= token_length(message.to_json, model_name, llm: @llm)
+
+          proceed
+        end
+      end
+
+      def summarize_messages
+        history = [@summary, @messages.to_json].compact.join("\n")
+        partitions = [history[0, history.size / 2], history[history.size / 2, history.size]]
+
+        @summary = partitions.map { |messages| @llm.summarize(text: messages.to_json) }.join("\n")
+
+        @messages = [@messages.last]
+      end
+
+      def partition_messages
+      end
+
+      def model_name
+        @llm.class::DEFAULTS[:chat_completion_model_name]
+      end
+
+      def token_length(content, model_name, options)
+        @llm.class::LENGTH_VALIDATOR.token_length(content, model_name, options)
+      end
+    end
+  end
+end

data/lib/langchain/conversation/message.rb

@@ -0,0 +1,48 @@
+# frozen_string_literal: true
+
+module Langchain
+  class Conversation
+    class Message
+      attr_reader :content
+
+      ROLE_MAPPING = {
+        context: "system",
+        prompt: "user",
+        response: "assistant"
+      }
+
+      def initialize(content)
+        @content = content
+      end
+
+      def role
+        ROLE_MAPPING[type]
+      end
+
+      def to_s
+        content
+      end
+
+      def to_h
+        {
+          role: role,
+          content: content
+        }
+      end
+
+      def ==(other)
+        to_json == other.to_json
+      end
+
+      def to_json(options = {})
+        to_h.to_json
+      end
+
+      private
+
+      def type
+        self.class.to_s.split("::").last.downcase.to_sym
+      end
+    end
+  end
+end

data/lib/langchain/{human_message.rb → conversation/prompt.rb}

@@ -1,9 +1,8 @@
 # frozen_string_literal: true
 
 module Langchain
-  class HumanMessage < Message
-    def type
-      "human"
+  class Conversation
+    class Prompt < Message
     end
   end
 end

data/lib/langchain/{system_message.rb → conversation/response.rb}

@@ -1,9 +1,8 @@
 # frozen_string_literal: true
 
 module Langchain
-  class SystemMessage < Message
-    def type
-      "system"
+  class Conversation
+    class Response < Message
     end
   end
 end

data/lib/langchain/conversation.rb

@@ -28,7 +28,7 @@ module Langchain
       @llm = llm
       @context = nil
       @examples = []
-      @memory = ConversationMemory.new(
+      @memory = ::Langchain::Conversation::Memory.new(
         llm: llm,
         messages: options.delete(:messages) || [],
         strategy: options.delete(:memory_strategy)
@@ -44,48 +44,47 @@ module Langchain
     # 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 SystemMessage.new(message)
+      @memory.set_context ::Langchain::Conversation::Context.new(message)
     end
 
     # Add examples to the conversation. Used to give the model a sense of the conversation.
-    # @param examples [Array<AIMessage|HumanMessage>] The examples to add to the conversation
+    # @param examples [Array<Prompt|Response>] 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 [AIMessage] The response from the model
+    # @return [Response] The response from the model
     def message(message)
-      human_message = HumanMessage.new(message)
-      @memory.append_message(human_message)
-      ai_message = llm_response(human_message)
+      @memory.append_message ::Langchain::Conversation::Prompt.new(message)
+      ai_message = ::Langchain::Conversation::Response.new(llm_response.chat_completion)
       @memory.append_message(ai_message)
       ai_message
     end
 
     # Messages from conversation memory
-    # @return [Array<AIMessage|HumanMessage>] The messages from the conversation memory
+    # @return [Array<Prompt|Response>] The messages from the conversation memory
     def messages
       @memory.messages
     end
 
     # Context from conversation memory
-    # @return [SystemMessage] Context from conversation memory
+    # @return [Context] Context from conversation memory
     def context
       @memory.context
     end
 
     # Examples from conversation memory
-    # @return [Array<AIMessage|HumanMessage>] Examples from the conversation memory
+    # @return [Array<Prompt|Response>] Examples from the conversation memory
     def examples
       @memory.examples
     end
 
     private
 
-    def llm_response(prompt)
-      @llm.chat(messages: @memory.messages, context: @memory.context, examples: @memory.examples, **@options, &@block)
+    def llm_response
+      @llm.chat(messages: @memory.messages.map(&:to_h), context: @memory.context&.to_s, examples: @memory.examples.map(&:to_h), **@options, &@block)
     rescue Langchain::Utils::TokenLength::TokenLimitExceeded => exception
       @memory.reduce_messages(exception)
       retry

data/lib/langchain/llm/ai21.rb

@@ -8,7 +8,7 @@ module Langchain::LLM
   #   gem "ai21", "~> 0.2.1"
   #
   # Usage:
-  #     ai21 = Langchain::LLM::AI21.new(api_key:)
+  #     ai21 = Langchain::LLM::AI21.new(api_key: ENV["AI21_API_KEY"])
   #
   class AI21 < Base
     DEFAULTS = {
@@ -30,7 +30,7 @@ module Langchain::LLM
     #
     # @param prompt [String] The prompt to generate a completion for
     # @param params [Hash] The parameters to pass to the API
-    # @return [String] The completion
+    # @return [Langchain::LLM::AI21Response] The completion
     #
     def complete(prompt:, **params)
       parameters = complete_parameters params
@@ -38,7 +38,7 @@ module Langchain::LLM
       parameters[:maxTokens] = LENGTH_VALIDATOR.validate_max_tokens!(prompt, parameters[:model], client)
 
       response = client.complete(prompt, parameters)
-      response.dig(:completions, 0, :data, :text)
+      Langchain::LLM::AI21Response.new response, model: parameters[:model]
     end
 
     #
@@ -51,6 +51,7 @@ module Langchain::LLM
     def summarize(text:, **params)
       response = client.summarize(text, "TEXT", params)
       response.dig(:summary)
+      # Should we update this to also return a Langchain::LLM::AI21Response?
     end
 
     private

data/lib/langchain/llm/anthropic.rb

@@ -8,7 +8,7 @@ module Langchain::LLM
   #   gem "anthropic", "~> 0.1.0"
   #
   # Usage:
-  #     anthorpic = Langchain::LLM::Anthropic.new(api_key:)
+  #     anthorpic = Langchain::LLM::Anthropic.new(api_key: ENV["ANTHROPIC_API_KEY"])
   #
   class Anthropic < Base
     DEFAULTS = {
@@ -32,7 +32,7 @@ module Langchain::LLM
     #
     # @param prompt [String] The prompt to generate a completion for
     # @param params [Hash] extra parameters passed to Anthropic::Client#complete
-    # @return [String] The completion
+    # @return [Langchain::LLM::AnthropicResponse] The completion
     #
     def complete(prompt:, **params)
       parameters = compose_parameters @defaults[:completion_model_name], params
@@ -43,7 +43,7 @@ module Langchain::LLM
       # parameters[:max_tokens_to_sample] = validate_max_tokens(prompt, parameters[:completion_model_name])
 
       response = client.complete(parameters: parameters)
-      response.dig("completion")
+      Langchain::LLM::AnthropicResponse.new(response)
     end
 
     private

data/lib/langchain/llm/cohere.rb

@@ -8,7 +8,7 @@ module Langchain::LLM
   #     gem "cohere-ruby", "~> 0.9.6"
   #
   # Usage:
-  #     cohere = Langchain::LLM::Cohere.new(api_key: "YOUR_API_KEY")
+  #     cohere = Langchain::LLM::Cohere.new(api_key: ENV["COHERE_API_KEY"])
   #
   class Cohere < Base
     DEFAULTS = {
@@ -30,14 +30,15 @@ module Langchain::LLM
     # Generate an embedding for a given text
     #
     # @param text [String] The text to generate an embedding for
-    # @return [Hash] The embedding
+    # @return [Langchain::LLM::CohereResponse] Response object
     #
     def embed(text:)
       response = client.embed(
         texts: [text],
         model: @defaults[:embeddings_model_name]
       )
-      response.dig("embeddings").first
+
+      Langchain::LLM::CohereResponse.new response, model: @defaults[:embeddings_model_name]
     end
 
     #
@@ -45,7 +46,7 @@ module Langchain::LLM
     #
     # @param prompt [String] The prompt to generate a completion for
     # @param params[:stop_sequences]
-    # @return [Hash] The completion
+    # @return [Langchain::LLM::CohereResponse] Response object
     #
     def complete(prompt:, **params)
       default_params = {
@@ -64,13 +65,13 @@ module Langchain::LLM
       default_params[:max_tokens] = Langchain::Utils::TokenLength::CohereValidator.validate_max_tokens!(prompt, default_params[:model], client)
 
       response = client.generate(**default_params)
-      response.dig("generations").first.dig("text")
+      Langchain::LLM::CohereResponse.new response, model: @defaults[:completion_model_name]
     end
 
     # Cohere does not have a dedicated chat endpoint, so instead we call `complete()`
     def chat(...)
       response_text = complete(...)
-      Langchain::AIMessage.new(response_text)
+      ::Langchain::Conversation::Response.new(response_text)
     end
 
     # Generate a summary in English for a given text