boxcars 0.6.3 → 0.6.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 8c7137b49fcea80018efc8f6dc19361fa02b74a575f7e71218784986b366c350
-  data.tar.gz: 9b2bba4216ab4f50024d158cc7b18977569fa7b9bada1a4734fb24f7b9e4a3be
+  metadata.gz: 164d55b95e134a691b4525f778a83f6d06cdaa1948028b87371cb6cb8a16fd43
+  data.tar.gz: 2707ac063bbc1127d75398be8770812ae258447ba021c0b9527cd3fbf113ff02
 SHA512:
-  metadata.gz: 18b58fc68b3837e2f8c7ca1e77e1daf49655843903d39a355233ff1d8c7a68caa885530ba08bcfc3842189b070c547ce152e13ee1a1b5cdb4cdc0424fe8e8ddc
-  data.tar.gz: e6240effad988b8cb819070ea37f54eab03dd99b32eff0caa07aa167e0f854492c12843d781e9614f429ffd3a311f426bc58c9569af22f7424bd9b65f60244fa
+  metadata.gz: 8b0d7909ef7c049d7c393b646abd072c01dd4b295a7af909bbe2f85fe48b8425c946f956440548fa3902182c636d3e57e81279332d331a9d7c4df4181cd4e4d5
+  data.tar.gz: ed1c903fb5575ae563ccab8adb0e80cfed9c00e529a5850fc7b89acaf165eecaf42e69cdd548cdec3622855645b5bf88591b3d6a6cb1af47bfbe97d5d6d16ad9

data/.ruby-version

@@ -1 +1 @@
-3.3.3
+3.3.5

data/CHANGELOG.md

@@ -1,5 +1,29 @@
 # Changelog
 
+## [v0.6.4](https://github.com/BoxcarsAI/boxcars/tree/v0.6.4) (2024-07-27)
+
+[Full Changelog](https://github.com/BoxcarsAI/boxcars/compare/v0.6.3...v0.6.4)
+
+**Merged pull requests:**
+
+- Add Ollama Engine [\#200](https://github.com/BoxcarsAI/boxcars/pull/200) ([francis](https://github.com/francis))
+
+## [v0.6.3](https://github.com/BoxcarsAI/boxcars/tree/v0.6.3) (2024-07-26)
+
+[Full Changelog](https://github.com/BoxcarsAI/boxcars/compare/v0.6.2...v0.6.3)
+
+**Merged pull requests:**
+
+- Add Groq engine [\#199](https://github.com/BoxcarsAI/boxcars/pull/199) ([francis](https://github.com/francis))
+
+## [v0.6.2](https://github.com/BoxcarsAI/boxcars/tree/v0.6.2) (2024-07-24)
+
+[Full Changelog](https://github.com/BoxcarsAI/boxcars/compare/v0.6.1...v0.6.2)
+
+**Merged pull requests:**
+
+- add flag for symbolizing JSON Engine Boxcar results [\#198](https://github.com/BoxcarsAI/boxcars/pull/198) ([francis](https://github.com/francis))
+
 ## [v0.6.1](https://github.com/BoxcarsAI/boxcars/tree/v0.6.1) (2024-07-19)
 
 [Full Changelog](https://github.com/BoxcarsAI/boxcars/compare/v0.5.1...v0.6.1)

data/Gemfile

@@ -15,13 +15,13 @@ gem "sqlite3", "~> 1.7"
 
 gem "async", "~>1.32.1"
 
-gem "activerecord", "~> 7.0"
+gem "activerecord", "~> 7.1"
 
 gem "github_changelog_generator", "~> 1.16"
 
 gem "faraday-retry", "~> 2.0"
 
-gem "activesupport", "~> 7.0"
+gem "activesupport", "~> 7.1"
 
 gem "rest-client", "~> 2.1"
 
@@ -32,9 +32,9 @@ gem "pgvector", "~> 0.2.2"
 
 group :development, :test do
   gem "rspec", "~> 3.13"
-  gem "rubocop", "~> 1.64"
-  gem "vcr", "~> 6.2.0"
+  gem "rubocop", "~> 1.66"
+  gem "vcr", "~> 6.3.1"
   gem "webmock", "~> 3.23.1"
   gem "rubocop-rake", "~> 0.6.0"
-  gem "rubocop-rspec", "~> 3.0"
+  gem "rubocop-rspec", "~> 3.1"
 end

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    boxcars (0.6.3)
+    boxcars (0.6.5)
       anthropic (~> 0.1)
       google_search_results (~> 2.2)
       gpt4all (~> 0.0.4)
@@ -13,13 +13,13 @@ PATH
 GEM
   remote: https://rubygems.org/
   specs:
-    activemodel (7.1.3.4)
-      activesupport (= 7.1.3.4)
-    activerecord (7.1.3.4)
-      activemodel (= 7.1.3.4)
-      activesupport (= 7.1.3.4)
+    activemodel (7.1.4)
+      activesupport (= 7.1.4)
+    activerecord (7.1.4)
+      activemodel (= 7.1.4)
+      activesupport (= 7.1.4)
       timeout (>= 0.4.0)
-    activesupport (7.1.3.4)
+    activesupport (7.1.4)
       base64
       bigdecimal
       concurrent-ruby (~> 1.0, >= 1.0.2)
@@ -57,7 +57,7 @@ GEM
       async (>= 1.25)
     base64 (0.2.0)
     bigdecimal (3.1.8)
-    concurrent-ruby (1.3.3)
+    concurrent-ruby (1.3.4)
     connection_pool (2.4.1)
     console (1.27.0)
       fiber-annotation
@@ -71,7 +71,7 @@ GEM
       reline (>= 0.3.8)
     diff-lcs (1.5.1)
     domain_name (0.6.20240107)
-    dotenv (3.1.2)
+    dotenv (3.1.4)
     drb (2.2.1)
     event_stream_parser (1.0.0)
     faraday (2.10.0)
@@ -120,7 +120,7 @@ GEM
     mime-types (3.5.2)
       mime-types-data (~> 3.2015)
     mime-types-data (3.2024.0702)
-    minitest (5.24.1)
+    minitest (5.25.1)
     multi_json (1.15.0)
     multipart-post (2.4.1)
     mutex_m (0.2.0)
@@ -128,19 +128,19 @@ GEM
       uri
     netrc (0.11.0)
     nio4r (2.7.3)
-    nokogiri (1.16.6-arm64-darwin)
+    nokogiri (1.16.7-arm64-darwin)
       racc (~> 1.4)
-    nokogiri (1.16.6-x86_64-linux)
+    nokogiri (1.16.7-x86_64-linux)
       racc (~> 1.4)
     octokit (4.25.1)
       faraday (>= 1, < 3)
       sawyer (~> 0.9)
     os (1.1.4)
-    parallel (1.25.1)
-    parser (3.3.4.0)
+    parallel (1.26.3)
+    parser (3.3.5.0)
       ast (~> 2.4.1)
       racc
-    pg (1.5.6)
+    pg (1.5.8)
     pgvector (0.2.2)
     protocol-hpack (1.4.3)
     protocol-http (0.26.8)
@@ -152,7 +152,7 @@ GEM
     psych (5.1.2)
       stringio
     public_suffix (6.0.0)
-    racc (1.8.0)
+    racc (1.8.1)
     rainbow (3.1.1)
     rake (13.2.1)
     rdoc (6.7.0)
@@ -165,8 +165,7 @@ GEM
       http-cookie (>= 1.0.2, < 2.0)
       mime-types (>= 1.16, < 4.0)
       netrc (~> 0.8)
-    rexml (3.3.2)
-      strscan
+    rexml (3.3.8)
     rspec (3.13.0)
       rspec-core (~> 3.13.0)
       rspec-expectations (~> 3.13.0)
@@ -180,22 +179,21 @@ GEM
       diff-lcs (>= 1.2.0, < 2.0)
       rspec-support (~> 3.13.0)
     rspec-support (3.13.1)
-    rubocop (1.65.0)
+    rubocop (1.66.1)
       json (~> 2.3)
       language_server-protocol (>= 3.17.0)
       parallel (~> 1.10)
       parser (>= 3.3.0.2)
       rainbow (>= 2.2.2, < 4.0)
       regexp_parser (>= 2.4, < 3.0)
-      rexml (>= 3.2.5, < 4.0)
-      rubocop-ast (>= 1.31.1, < 2.0)
+      rubocop-ast (>= 1.32.2, < 2.0)
       ruby-progressbar (~> 1.7)
       unicode-display_width (>= 2.4.0, < 3.0)
-    rubocop-ast (1.31.3)
+    rubocop-ast (1.32.3)
       parser (>= 3.3.1.0)
     rubocop-rake (0.6.0)
       rubocop (~> 1.0)
-    rubocop-rspec (3.0.3)
+    rubocop-rspec (3.1.0)
       rubocop (~> 1.61)
     ruby-openai (7.1.0)
       event_stream_parser (>= 0.3.0, < 2.0.0)
@@ -209,7 +207,6 @@ GEM
     sqlite3 (1.7.3-x86_64-linux)
     stringio (3.1.1)
     strings-ansi (0.2.0)
-    strscan (3.1.0)
     timeout (0.4.1)
     timers (4.3.5)
     traces (0.11.1)
@@ -222,9 +219,10 @@ GEM
     tty-screen (0.8.2)
     tzinfo (2.0.6)
       concurrent-ruby (~> 1.0)
-    unicode-display_width (2.5.0)
+    unicode-display_width (2.6.0)
     uri (0.13.0)
-    vcr (6.2.0)
+    vcr (6.3.1)
+      base64
     webmock (3.23.1)
       addressable (>= 2.8.0)
       crack (>= 0.3.2)
@@ -233,11 +231,12 @@ GEM
 PLATFORMS
   arm64-darwin-22
   arm64-darwin-23
+  arm64-darwin-24
   x86_64-linux
 
 DEPENDENCIES
-  activerecord (~> 7.0)
-  activesupport (~> 7.0)
+  activerecord (~> 7.1)
+  activesupport (~> 7.1)
   async (~> 1.32.1)
   boxcars!
   debug (~> 1.9)
@@ -250,11 +249,11 @@ DEPENDENCIES
   rake (~> 13.2)
   rest-client (~> 2.1)
   rspec (~> 3.13)
-  rubocop (~> 1.64)
+  rubocop (~> 1.66)
   rubocop-rake (~> 0.6.0)
-  rubocop-rspec (~> 3.0)
+  rubocop-rspec (~> 3.1)
   sqlite3 (~> 1.7)
-  vcr (~> 6.2.0)
+  vcr (~> 6.3.1)
   webmock (~> 3.23.1)
 
 BUNDLED WITH

data/lib/boxcars/boxcar/json_engine_boxcar.rb

@@ -50,12 +50,14 @@ module Boxcars
     # @param engine_output [String] The output from the engine.
     # @return [Result] The result.
     def get_answer(engine_output)
-      # sometimes the LLM adds text in front of the JSON output, so let's strip it here
-      json_start = engine_output.index("{")
-      json_end = engine_output.rindex("}")
-      extract_answer(JSON.parse(engine_output[json_start..json_end], symbolize_names: symbolize))
+      json_string = extract_json(engine_output)
+      reply = JSON.parse(json_string, symbolize_names: symbolize)
+      Result.new(status: :ok, answer: reply, explanation: reply)
+    rescue JSON::ParserError => e
+      Boxcars.debug "JSON: #{engine_output}", :red
+      Result.from_error("JSON parsing error: #{e.message}")
     rescue StandardError => e
-      Result.from_error("Error: #{e.message}:\n#{engine_output}")
+      Result.from_error("Unexpected error: #{e.message}")
     end
 
     # get answer from parsed JSON
@@ -63,14 +65,29 @@ module Boxcars
     # @return [Result] The result.
     def extract_answer(data)
       reply = data
+      Result.new(status: :ok, answer: reply, explanation: reply)
+    end
+
+    private
+
+    def extract_json(text)
+      # Escape control characters (U+0000 to U+001F)
+      text = text.gsub(/[\u0000-\u001F]/, '')
+      # first strip hidden characters
+      # text = text.encode('UTF-8', invalid: :replace, undef: :replace, replace: '')
+
+      # sometimes the LLM adds text in front of the JSON output, so let's strip it here
+      json_start = text.index("{")
+      json_end = text.rindex("}")
+      text[json_start..json_end]
+    end
+
+    def extract_json2(text)
+      # Match the outermost JSON object
+      match = text.match(/\{(?:[^{}]|\{(?:[^{}]|\{[^{}]*\})*\})*\}/)
+      raise StandardError, "No valid JSON object found in the output" unless match
 
-      if reply.present?
-        Result.new(status: :ok, answer: reply, explanation: reply)
-      else
-        # we have an unexpected output from the engine
-        Result.new(status: :error, answer: nil,
-                   explanation: "You gave me an improperly formatted answer. I was expecting a valid reply.")
-      end
+      match[0]
     end
   end
 end

data/lib/boxcars/boxcar.rb

@@ -166,7 +166,7 @@ module Boxcars
         output = call(inputs: inputs)
       rescue StandardError => e
         Boxcars.error "Error in #{name} boxcar#call: #{e}\nbt:#{e.backtrace[0..5].join("\n   ")}", :red
-        Boxcars.error("Response Body: #{e.response[:body]}", :red) if e.respond_to?(:response)
+        Boxcars.error("Response Body: #{e.response[:body]}", :red) if e.respond_to?(:response) && e.response.present?
         raise e
       end
       validate_outputs(outputs: output.keys)

data/lib/boxcars/engine/cohere.rb

@@ -90,21 +90,6 @@ module Boxcars
       llm_params
     end
 
-    # Get generation informaton
-    # @param sub_choices [Array<Hash>] The choices to get generation info for.
-    # @return [Array<Generation>] The generation information.
-    def generation_info(sub_choices)
-      sub_choices.map do |choice|
-        Generation.new(
-          text: choice["completion"],
-          generation_info: {
-            finish_reason: choice.fetch("stop_reason", nil),
-            logprobs: choice.fetch("logprobs", nil)
-          }
-        )
-      end
-    end
-
     # make sure we got a valid response
     # @param response [Hash] The response to check.
     # @param must_haves [Array<String>] The keys that must be in the response. Defaults to %w[choices].
@@ -124,45 +109,11 @@ module Boxcars
       end
     end
 
-    # Call out to OpenAI's endpoint with k unique prompts.
-    # @param prompts [Array<String>] The prompts to pass into the model.
-    # @param inputs [Array<String>] The inputs to subsitite into the prompt.
-    # @param stop [Array<String>] Optional list of stop words to use when generating.
-    # @return [EngineResult] The full engine output.
-    def generate(prompts:, stop: nil)
-      params = {}
-      params[:stop] = stop if stop
-      choices = []
-      # Get the token usage from the response.
-      # Includes prompt, completion, and total tokens used.
-      prompts.each_slice(batch_size) do |sub_prompts|
-        sub_prompts.each do |sprompts, inputs|
-          response = client(prompt: sprompts, inputs: inputs, **params)
-          check_response(response)
-          choices << response
-        end
-      end
-
-      n = params.fetch(:n, 1)
-      generations = []
-      prompts.each_with_index do |_prompt, i|
-        sub_choices = choices[i * n, (i + 1) * n]
-        generations.push(generation_info(sub_choices))
-      end
-      EngineResult.new(generations: generations, engine_output: { token_usage: {} })
-    end
-    # rubocop:enable Metrics/AbcSize
-
     # the engine type
     def engine_type
       "claude"
     end
 
-    # calculate the number of tokens used
-    def get_num_tokens(text:)
-      text.split.length # TODO: hook up to token counting gem
-    end
-
     # lookup the context size for a model by name
     # @param modelname [String] The name of the model to lookup.
     def modelname_to_contextsize(_modelname)

data/lib/boxcars/engine/groq.rb

@@ -71,8 +71,8 @@ module Boxcars
       prompt = Prompt.new(template: question)
       response = client(prompt: prompt, **kwargs)
       raise Error, "Groq: No response from API" unless response
-      raise Error, "Groq: #{response['error']}" if response["error"]
 
+      check_response(response)
       answer = response["choices"].map { |c| c.dig("message", "content") || c["text"] }.join("\n").strip
       puts answer
       answer
@@ -83,31 +83,16 @@ module Boxcars
       groq_parmas
     end
 
-    # Get generation informaton
-    # @param sub_choices [Array<Hash>] The choices to get generation info for.
-    # @return [Array<Generation>] The generation information.
-    def generation_info(sub_choices)
-      sub_choices.map do |choice|
-        Generation.new(
-          text: choice.dig("message", "content") || choice["text"],
-          generation_info: {
-            finish_reason: choice.fetch("finish_reason", nil),
-            logprobs: choice.fetch("logprobs", nil)
-          }
-        )
-      end
-    end
-
     # make sure we got a valid response
     # @param response [Hash] The response to check.
     # @param must_haves [Array<String>] The keys that must be in the response. Defaults to %w[choices].
     # @raise [KeyError] if there is an issue with the access token.
     # @raise [ValueError] if the response is not valid.
     def check_response(response, must_haves: %w[choices])
-      if response['error']
+      if response['error'].is_a?(Hash)
         code = response.dig('error', 'code')
         msg = response.dig('error', 'message') || 'unknown error'
-        raise KeyError, "OPENAI_ACCESS_TOKEN not valid" if code == 'invalid_api_key'
+        raise KeyError, "GROQ_API_TOKEN not valid" if code == 'invalid_api_key'
 
         raise ValueError, "Groq error: #{msg}"
       end
@@ -117,58 +102,20 @@ module Boxcars
       end
     end
 
-    # Call out to Groq's endpoint with k unique prompts.
-    # @param prompts [Array<String>] The prompts to pass into the model.
-    # @param inputs [Array<String>] The inputs to subsitite into the prompt.
-    # @param stop [Array<String>] Optional list of stop words to use when generating.
-    # @return [EngineResult] The full engine output.
-    def generate(prompts:, stop: nil)
-      params = {}
-      params[:stop] = stop if stop
-      choices = []
-      token_usage = {}
-      # Get the token usage from the response.
-      # Includes prompt, completion, and total tokens used.
-      inkeys = %w[completion_tokens prompt_tokens total_tokens].freeze
-      prompts.each_slice(batch_size) do |sub_prompts|
-        sub_prompts.each do |sprompts, inputs|
-          response = client(prompt: sprompts, inputs: inputs, **params)
-          check_response(response)
-          choices.concat(response["choices"])
-          usage_keys = inkeys & response["usage"].keys
-          usage_keys.each { |key| token_usage[key] = token_usage[key].to_i + response["usage"][key] }
-        end
-      end
-
-      n = params.fetch(:n, 1)
-      generations = []
-      prompts.each_with_index do |_prompt, i|
-        sub_choices = choices[i * n, (i + 1) * n]
-        generations.push(generation_info(sub_choices))
-      end
-      EngineResult.new(generations: generations, engine_output: { token_usage: token_usage })
+    # the engine type
+    def engine_type
+      "groq"
     end
-    # rubocop:enable Metrics/AbcSize
-  end
-
-  # the engine type
-  def engine_type
-    "groq"
-  end
 
-  # calculate the number of tokens used
-  def get_num_tokens(text:)
-    text.split.length # TODO: hook up to token counting gem
-  end
-
-  # Calculate the maximum number of tokens possible to generate for a prompt.
-  # @param prompt_text [String] The prompt text to use.
-  # @return [Integer] the number of tokens possible to generate.
-  def max_tokens_for_prompt(prompt_text)
-    num_tokens = get_num_tokens(prompt_text)
+    # Calculate the maximum number of tokens possible to generate for a prompt.
+    # @param prompt_text [String] The prompt text to use.
+    # @return [Integer] the number of tokens possible to generate.
+    def max_tokens_for_prompt(prompt_text)
+      num_tokens = get_num_tokens(prompt_text)
 
-    # get max context size for model by name
-    max_size = 8096
-    max_size - num_tokens
+      # get max context size for model by name
+      max_size = 8096
+      max_size - num_tokens
+    end
   end
 end

data/lib/boxcars/engine/ollama.rb

@@ -0,0 +1,80 @@
+# frozen_string_literal: true
+
+# Boxcars is a framework for running a series of tools to get an answer to a question.
+module Boxcars
+  # A engine that uses local GPT4All API.
+  class Ollama < Engine
+    attr_reader :prompts, :model_kwargs, :batch_size, :ollama_params
+
+    # The default parameters to use when asking the engine.
+    DEFAULT_PARAMS = {
+      model: "llama3",
+      temperature: 0.1,
+      max_tokens: 4096
+    }.freeze
+
+    # the default name of the engine
+    DEFAULT_NAME = "Ollama engine"
+    # the default description of the engine
+    DEFAULT_DESCRIPTION = "useful for when you need to use local AI to answer questions. " \
+                          "You should ask targeted questions"
+
+    # A engine is a container for a single tool to run.
+    # @param name [String] The name of the engine. Defaults to "OpenAI engine".
+    # @param description [String] A description of the engine. Defaults to:
+    #        useful for when you need to use AI to answer questions. You should ask targeted questions".
+    # @param prompts [Array<String>] The prompts to use when asking the engine. Defaults to [].
+    # @param batch_size [Integer] The number of prompts to send to the engine at once. Defaults to 2.
+    def initialize(name: DEFAULT_NAME, description: DEFAULT_DESCRIPTION, prompts: [], batch_size: 2, **kwargs)
+      @ollama_params = DEFAULT_PARAMS.merge(kwargs)
+      @prompts = prompts
+      @batch_size = batch_size
+      super(description: description, name: name)
+    end
+
+    # Get the OpenAI API client
+    # @param groq_api_key [String] The access token to use when asking the engine.
+    #   Defaults to Boxcars.configuration.groq_api_key
+    # @return [OpenAI::Client] The OpenAI API gem client.
+    def self.open_ai_client
+      ::OpenAI::Client.new(uri_base: "http://localhost:11434")
+    end
+
+    def conversation_model?(_model)
+      true
+    end
+
+    # Get an answer from the engine.
+    # @param prompt [String] The prompt to use when asking the engine.
+    # @param groq_api_key [String] The access token to use when asking the engine.
+    #   Defaults to Boxcars.configuration.groq_api_key.
+    # @param kwargs [Hash] Additional parameters to pass to the engine if wanted.
+    def client(prompt:, inputs: {}, **kwargs)
+      clnt = Ollama.open_ai_client
+      params = ollama_params.merge(kwargs)
+      prompt = prompt.first if prompt.is_a?(Array)
+      params = prompt.as_messages(inputs).merge(params)
+      if Boxcars.configuration.log_prompts
+        Boxcars.debug(params[:messages].last(2).map { |p| ">>>>>> Role: #{p[:role]} <<<<<<\n#{p[:content]}" }.join("\n"), :cyan)
+      end
+      ans = clnt.chat(parameters: params)
+      ans['choices'].pluck('message').pluck('content').join("\n")
+    rescue => e
+      Boxcars.error(e, :red)
+      raise
+    end
+
+    # get an answer from the engine for a question.
+    # @param question [String] The question to ask the engine.
+    # @param kwargs [Hash] Additional parameters to pass to the engine if wanted.
+    def run(question, **kwargs)
+      prompt = Prompt.new(template: question)
+      answer = client(prompt: prompt, **kwargs)
+      raise Error, "Ollama: No response from API" unless answer
+
+      # raise Error, "Ollama: #{response['error']}" if response["error"]
+      Boxcars.debug("Answer: #{answer}", :cyan)
+      answer
+    end
+  end
+end

data/lib/boxcars/engine/openai.rb

@@ -28,6 +28,11 @@ module Boxcars
     # @param batch_size [Integer] The number of prompts to send to the engine at once. Defaults to 20.
     def initialize(name: DEFAULT_NAME, description: DEFAULT_DESCRIPTION, prompts: [], batch_size: 20, **kwargs)
       @open_ai_params = DEFAULT_PARAMS.merge(kwargs)
+      if @open_ai_params[:model] =~ /^o/ && @open_ai_params[:max_tokens].present?
+        @open_ai_params[:max_completion_tokens] = @open_ai_params.delete(:max_tokens)
+        @open_ai_params.delete(:temperature)
+      end
+
       @prompts = prompts
       @batch_size = batch_size
       super(description: description, name: name)
@@ -44,7 +49,7 @@ module Boxcars
     end
 
     def conversation_model?(model)
-      !!(model =~ /(^gpt-4)|(-turbo\b)/)
+      !!(model =~ /(^gpt-4)|(-turbo\b)|(^o\d)/)
     end
 
     # Get an answer from the engine.
@@ -57,6 +62,10 @@ module Boxcars
       params = open_ai_params.merge(kwargs)
       if conversation_model?(params[:model])
         prompt = prompt.first if prompt.is_a?(Array)
+        if params[:model] =~ /^o/
+          params.delete(:response_format)
+          params.delete(:stop)
+        end
         params = prompt.as_messages(inputs).merge(params)
         if Boxcars.configuration.log_prompts
           Boxcars.debug(params[:messages].last(2).map { |p| ">>>>>> Role: #{p[:role]} <<<<<<\n#{p[:content]}" }.join("\n"), :cyan)
@@ -88,21 +97,6 @@ module Boxcars
       open_ai_params
     end
 
-    # Get generation informaton
-    # @param sub_choices [Array<Hash>] The choices to get generation info for.
-    # @return [Array<Generation>] The generation information.
-    def generation_info(sub_choices)
-      sub_choices.map do |choice|
-        Generation.new(
-          text: choice.dig("message", "content") || choice["text"],
-          generation_info: {
-            finish_reason: choice.fetch("finish_reason", nil),
-            logprobs: choice.fetch("logprobs", nil)
-          }
-        )
-      end
-    end
-
     # make sure we got a valid response
     # @param response [Hash] The response to check.
     # @param must_haves [Array<String>] The keys that must be in the response. Defaults to %w[choices].
@@ -121,39 +115,6 @@ module Boxcars
         raise ValueError, "Expecting key #{key} in response" unless response.key?(key)
       end
     end
-
-    # Call out to OpenAI's endpoint with k unique prompts.
-    # @param prompts [Array<String>] The prompts to pass into the model.
-    # @param inputs [Array<String>] The inputs to subsitite into the prompt.
-    # @param stop [Array<String>] Optional list of stop words to use when generating.
-    # @return [EngineResult] The full engine output.
-    def generate(prompts:, stop: nil)
-      params = {}
-      params[:stop] = stop if stop
-      choices = []
-      token_usage = {}
-      # Get the token usage from the response.
-      # Includes prompt, completion, and total tokens used.
-      inkeys = %w[completion_tokens prompt_tokens total_tokens].freeze
-      prompts.each_slice(batch_size) do |sub_prompts|
-        sub_prompts.each do |sprompts, inputs|
-          response = client(prompt: sprompts, inputs: inputs, **params)
-          check_response(response)
-          choices.concat(response["choices"])
-          usage_keys = inkeys & response["usage"].keys
-          usage_keys.each { |key| token_usage[key] = token_usage[key].to_i + response["usage"][key] }
-        end
-      end
-
-      n = params.fetch(:n, 1)
-      generations = []
-      prompts.each_with_index do |_prompt, i|
-        sub_choices = choices[i * n, (i + 1) * n]
-        generations.push(generation_info(sub_choices))
-      end
-      EngineResult.new(generations: generations, engine_output: { token_usage: token_usage })
-    end
-    # rubocop:enable Metrics/AbcSize
   end
 
   # the engine type
@@ -161,11 +122,6 @@ module Boxcars
     "openai"
   end
 
-  # calculate the number of tokens used
-  def get_num_tokens(text:)
-    text.split.length # TODO: hook up to token counting gem
-  end
-
   # lookup the context size for a model by name
   # @param modelname [String] The name of the model to lookup.
   def modelname_to_contextsize(modelname)

data/lib/boxcars/engine/perplexityai.rb

@@ -8,9 +8,8 @@ module Boxcars
 
     # The default parameters to use when asking the engine.
     DEFAULT_PER_PARAMS = {
-      model: "llama-2-70b-chat",
-      temperature: 0.1,
-      max_tokens: 3200
+      model: "'llama-3-sonar-large-32k-online'",
+      temperature: 0.1
     }.freeze
 
     # the default name of the engine
@@ -32,28 +31,27 @@ module Boxcars
       super(description: description, name: name)
     end
 
-    def conversation_model?(model)
-      ["mistral-7b-instruct", "llama-2-13b-chat", "llama-2-70b-chat", "openhermes-2-mistral-7b"].include?(model)
+    def conversation_model?(_model)
+      true
     end
 
     def chat(parameters:)
-      url = URI("https://api.perplexity.ai/chat/completions")
-
-      http = Net::HTTP.new(url.host, url.port)
-      http.use_ssl = true
-
-      request = Net::HTTP::Post.new(url)
-      request["accept"] = 'application/json'
-      request["authorization"] = "Bearer #{ENV.fetch('PERPLEXITY_API_KEY')}"
-      request["content-type"] = 'application/json'
-      the_body = {
-        model: parameters[:model] || "mistral-7b-instruct",
-        messages: parameters[:messages]
-      }
-      request.body = the_body.to_json
-
-      response = http.request(request)
-      JSON.parse(response.read_body)
+      conn = Faraday.new(url: "https://api.perplexity.ai/chat/completions") do |faraday|
+        faraday.request :json
+        faraday.response :json
+        faraday.response :raise_error
+        # faraday.options.timeout = 180 # 3 minutes
+      end
+
+      response = conn.post do |req|
+        req.headers['Authorization'] = "Bearer #{ENV.fetch('PERPLEXITY_API_KEY')}"
+        req.body = {
+          model: parameters[:model],
+          messages: parameters[:messages]
+        }
+      end
+
+      response.body
     end
 
     # Get an answer from the engine.
@@ -64,7 +62,6 @@ module Boxcars
     def client(prompt:, inputs: {}, **kwargs)
       prompt = prompt.first if prompt.is_a?(Array)
       params = prompt.as_messages(inputs).merge(default_params).merge(kwargs)
-      params[:model] ||= "llama-2-70b-chat"
       if Boxcars.configuration.log_prompts
         Boxcars.debug(params[:messages].last(2).map { |p| ">>>>>> Role: #{p[:role]} <<<<<<\n#{p[:content]}" }.join("\n"), :cyan)
       end
@@ -90,21 +87,6 @@ module Boxcars
       perplexity_params
     end
 
-    # Get generation informaton
-    # @param sub_choices [Array<Hash>] The choices to get generation info for.
-    # @return [Array<Generation>] The generation information.
-    def generation_info(sub_choices)
-      sub_choices.map do |choice|
-        Generation.new(
-          text: choice.dig("message", "content") || choice["text"],
-          generation_info: {
-            finish_reason: choice.fetch("finish_reason", nil),
-            logprobs: choice.fetch("logprobs", nil)
-          }
-        )
-      end
-    end
-
     # make sure we got a valid response
     # @param response [Hash] The response to check.
     # @param must_haves [Array<String>] The keys that must be in the response. Defaults to %w[choices].
@@ -123,39 +105,6 @@ module Boxcars
         raise ValueError, "Expecting key #{key} in response" unless response.key?(key)
       end
     end
-
-    # Call out to OpenAI's endpoint with k unique prompts.
-    # @param prompts [Array<String>] The prompts to pass into the model.
-    # @param inputs [Array<String>] The inputs to subsitite into the prompt.
-    # @param stop [Array<String>] Optional list of stop words to use when generating.
-    # @return [EngineResult] The full engine output.
-    def generate(prompts:, stop: nil)
-      params = {}
-      params[:stop] = stop if stop
-      choices = []
-      token_usage = {}
-      # Get the token usage from the response.
-      # Includes prompt, completion, and total tokens used.
-      inkeys = %w[completion_tokens prompt_tokens total_tokens].freeze
-      prompts.each_slice(batch_size) do |sub_prompts|
-        sub_prompts.each do |sprompts, inputs|
-          response = client(prompt: sprompts, inputs: inputs, **params)
-          check_response(response)
-          choices.concat(response["choices"])
-          usage_keys = inkeys & response["usage"].keys
-          usage_keys.each { |key| token_usage[key] = token_usage[key].to_i + response["usage"][key] }
-        end
-      end
-
-      n = params.fetch(:n, 1)
-      generations = []
-      prompts.each_with_index do |_prompt, i|
-        sub_choices = choices[i * n, (i + 1) * n]
-        generations.push(generation_info(sub_choices))
-      end
-      EngineResult.new(generations: generations, engine_output: { token_usage: token_usage })
-    end
-    # rubocop:enable Metrics/AbcSize
   end
 
   # the engine type
@@ -168,29 +117,57 @@ module Boxcars
     text.split.length # TODO: hook up to token counting gem
   end
 
-  # lookup the context size for a model by name
-  # @param modelname [String] The name of the model to lookup.
-  def modelname_to_contextsize(modelname)
-    model_lookup = {
-      'text-davinci-003': 4097,
-      'text-curie-001': 2048,
-      'text-babbage-001': 2048,
-      'text-ada-001': 2048,
-      'code-davinci-002': 8000,
-      'code-cushman-001': 2048,
-      'gpt-3.5-turbo-1': 4096
-    }.freeze
-    model_lookup[modelname] || 4097
-  end
-
   # Calculate the maximum number of tokens possible to generate for a prompt.
   # @param prompt_text [String] The prompt text to use.
   # @return [Integer] the number of tokens possible to generate.
-  def max_tokens_for_prompt(prompt_text)
-    num_tokens = get_num_tokens(prompt_text)
+  def max_tokens_for_prompt(_prompt_text)
+    8096
+  end
 
-    # get max context size for model by name
-    max_size = modelname_to_contextsize(model_name)
-    max_size - num_tokens
+  # Get generation informaton
+  # @param sub_choices [Array<Hash>] The choices to get generation info for.
+  # @return [Array<Generation>] The generation information.
+  def generation_info(sub_choices)
+    sub_choices.map do |choice|
+      Generation.new(
+        text: choice.dig("message", "content") || choice["text"],
+        generation_info: {
+          finish_reason: choice.fetch("finish_reason", nil),
+          logprobs: choice.fetch("logprobs", nil)
+        }
+      )
+    end
+  end
+
+  # Call out to endpoint with k unique prompts.
+  # @param prompts [Array<String>] The prompts to pass into the model.
+  # @param inputs [Array<String>] The inputs to subsitite into the prompt.
+  # @param stop [Array<String>] Optional list of stop words to use when generating.
+  # @return [EngineResult] The full engine output.
+  def generate(prompts:, stop: nil)
+    params = {}
+    params[:stop] = stop if stop
+    choices = []
+    token_usage = {}
+    # Get the token usage from the response.
+    # Includes prompt, completion, and total tokens used.
+    inkeys = %w[completion_tokens prompt_tokens total_tokens].freeze
+    prompts.each_slice(batch_size) do |sub_prompts|
+      sub_prompts.each do |sprompts, inputs|
+        response = client(prompt: sprompts, inputs: inputs, **params)
+        check_response(response)
+        choices.concat(response["choices"])
+        usage_keys = inkeys & response["usage"].keys
+        usage_keys.each { |key| token_usage[key] = token_usage[key].to_i + response["usage"][key] }
+      end
+    end
+
+    n = params.fetch(:n, 1)
+    generations = []
+    prompts.each_with_index do |_prompt, i|
+      sub_choices = choices[i * n, (i + 1) * n]
+      generations.push(generation_info(sub_choices))
+    end
+    EngineResult.new(generations: generations, engine_output: { token_usage: token_usage })
   end
 end

data/lib/boxcars/engine.rb

@@ -16,6 +16,58 @@ module Boxcars
     def run(question)
       raise NotImplementedError
     end
+
+    # calculate the number of tokens used
+    def get_num_tokens(text:)
+      text.split.length # TODO: hook up to token counting gem
+    end
+
+    # Get generation informaton
+    # @param sub_choices [Array<Hash>] The choices to get generation info for.
+    # @return [Array<Generation>] The generation information.
+    def generation_info(sub_choices)
+      sub_choices.map do |choice|
+        Generation.new(
+          text: choice.dig("message", "content") || choice["text"],
+          generation_info: {
+            finish_reason: choice.fetch("finish_reason", nil),
+            logprobs: choice.fetch("logprobs", nil)
+          }
+        )
+      end
+    end
+
+    # Call out to OpenAI's endpoint with k unique prompts.
+    # @param prompts [Array<String>] The prompts to pass into the model.
+    # @param inputs [Array<String>] The inputs to subsitite into the prompt.
+    # @param stop [Array<String>] Optional list of stop words to use when generating.
+    # @return [EngineResult] The full engine output.
+    def generate(prompts:, stop: nil)
+      params = {}
+      params[:stop] = stop if stop
+      choices = []
+      token_usage = {}
+      # Get the token usage from the response.
+      # Includes prompt, completion, and total tokens used.
+      inkeys = %w[completion_tokens prompt_tokens total_tokens].freeze
+      prompts.each_slice(batch_size) do |sub_prompts|
+        sub_prompts.each do |sprompts, inputs|
+          response = client(prompt: sprompts, inputs: inputs, **params)
+          check_response(response)
+          choices.concat(response["choices"])
+          usage_keys = inkeys & response["usage"].keys
+          usage_keys.each { |key| token_usage[key] = token_usage[key].to_i + response["usage"][key] }
+        end
+      end
+
+      n = params.fetch(:n, 1)
+      generations = []
+      prompts.each_with_index do |_prompt, i|
+        sub_choices = choices[i * n, (i + 1) * n]
+        generations.push(generation_info(sub_choices))
+      end
+      EngineResult.new(generations: generations, engine_output: { token_usage: token_usage })
+    end
   end
 end
 
@@ -23,6 +75,7 @@ require "boxcars/engine/engine_result"
 require "boxcars/engine/anthropic"
 require "boxcars/engine/cohere"
 require "boxcars/engine/groq"
+require "boxcars/engine/ollama"
 require "boxcars/engine/openai"
 require "boxcars/engine/perplexityai"
 require "boxcars/engine/gpt4all_eng"

data/lib/boxcars/version.rb

@@ -2,5 +2,5 @@
 
 module Boxcars
   # The current version of the gem.
-  VERSION = "0.6.3"
+  VERSION = "0.6.5"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: boxcars
 version: !ruby/object:Gem::Version
-  version: 0.6.3
+  version: 0.6.5
 platform: ruby
 authors:
 - Francis Sullivan
@@ -9,7 +9,7 @@ authors:
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-07-26 00:00:00.000000000 Z
+date: 2024-10-04 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: anthropic
@@ -161,6 +161,7 @@ files:
 - lib/boxcars/engine/engine_result.rb
 - lib/boxcars/engine/gpt4all_eng.rb
 - lib/boxcars/engine/groq.rb
+- lib/boxcars/engine/ollama.rb
 - lib/boxcars/engine/openai.rb
 - lib/boxcars/engine/perplexityai.rb
 - lib/boxcars/generation.rb
@@ -218,7 +219,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.4.10
+rubygems_version: 3.5.20
 signing_key:
 specification_version: 4
 summary: Boxcars is a gem that enables you to create new systems with AI composability.