src: refactor + clean code using better types

src/builders/generic.cr

@@ -1,6 +0,0 @@
-
-module Builder
-  abstract class PromptGeneric
-    abstract def build(prompt : Prompt)
-  end
-end

src/config.cr

@@ -0,0 +1,14 @@
+
+require "./openai_config"
+
+class AppConfig
+  property input_file = STDIN
+  property input_file_path = ""
+  property output_file = STDOUT
+  property output_file_path = ""
+  property past_characters_limit = 1000
+  property use_color = true
+  property make_request = true
+  property verbose = false
+  property gpt_config = AIUtils::GptConfig.new
+end

src/context_builders/generic_builder.cr

@@ -0,0 +1,6 @@
+
+module ContextBuilder
+  abstract class GenericContextBuilder
+    abstract def build(prompt : Prompt)
+  end
+end

src/context_builders/openai_chat.cr

@@ -14,10 +14,12 @@ module Builder
 
     # skip prelude_zone
     # skip future_zone
-    def build(prompt : Prompt) : Chat
+    def build(prompt : Prompt, context_token_limit : UInt32) : Chat
       chat = [] of Message
 
-      token_limit = 2_700
+     # token_limit = 10_000
+      token_limit = (0.65 * context_token_limit).to_i
+      STDERR.puts "messages_token_limit = #{token_limit}"
       mandatory_token_count = (
         prompt.system_zone.token_count + 
         prompt.present_zone.token_count

src/context_builders/prompt_dir.cr

@@ -0,0 +1,45 @@
+
+require "colorize"
+
+require "../config"
+require "./generic_builder"
+
+module Builder
+  class PromptDir < PromptGeneric
+
+    SEPARATOR = "\n\n"
+
+    getter use_color : Bool
+    def initialize(@use_color)
+      Colorize.enabled = @use_color
+    end
+
+    def build(prompt : Prompt)
+      str = ""
+
+      str += prompt.prelude_zone.content.join(SEPARATOR)
+
+      if ! prompt.system_zone.content.empty?
+        str += "@@system".colorize(:yellow).to_s
+        str += prompt.system_zone.content.join(SEPARATOR)
+      end
+
+      if ! prompt.past_zone.content.empty?
+        str += "@@before".colorize(:yellow).to_s
+        str += prompt.past_zone.content.join(SEPARATOR)
+      end
+
+      if ! prompt.present_zone.content.empty?
+        # str += "@@current".colorize(:yellow).to_s
+        str += prompt.present_zone.content.join(SEPARATOR).colorize(:light_cyan).to_s
+      end
+
+      if ! prompt.future_zone.content.empty?
+        str += "@@after".colorize(:yellow).to_s
+        str += prompt.future_zone.content.join("\n")
+      end
+
+      str
+    end
+  end
+end

src/context_builders/prompt_string.cr

@@ -1,10 +1,10 @@
 
 require "colorize"
 
-require "./generic"
+require "./generic_builder"
 
-module Builder
+module ContextBuilder
-  class PromptString < PromptGeneric
+  class PromptString < GenericContextBuilder
 
     SEPARATOR = "\n\n"
 

src/main.cr

@@ -3,15 +3,15 @@ require "pretty_print"
 require "openai"
 require "spinner"
 
-require "./zone"
+require "./prompts/zone"
 require "./parsers/prompt_string"
-require "./builders/prompt_string"
+require "./context_builders/prompt_string"
-require "./builders/openai_chat"
+# require "./context_builders/prompt_dir"
-require "./builders/openai_insert"
+require "./context_builders/openai_chat"
+require "./context_builders/openai_insert"
 require "./storyteller"
 
 storyteller = Storyteller.new()
 storyteller.prepare()
 storyteller.run()
 
-

src/modifiers/readme.md

@@ -0,0 +1,11 @@
+
+Des classes qui modifient un prompt d'apres des contraintes
+
+    truncate(max_tokens) # le prompt aux X derniers tokens
+
+    summarize() # réduis la longueur du prompt
+
+    worldinfo() # introduis les éléments du monde
+    
+    
+

src/openai_config.cr

@@ -0,0 +1,64 @@
+
+
+module AIUtils
+  enum OpenAIModel
+    Gpt_4 
+    Gpt_3_5_Turbo 
+    Gpt_3_5_Turbo_16k 
+
+    def to_s
+      vals = [
+        "gpt-4",
+       "gpt-3.5-turbo",
+       "gpt-3.5-turbo-16k"
+       ]
+      vals[self.value]
+    end
+
+    def self.from_s?(str_model : String) : OpenAIModel?
+      return OpenAIModel.values.find do |openai_model|
+       openai_model.to_s.downcase == str_model.downcase
+      end
+    end
+  end
+
+  enum OpenAIMode
+    Chat
+    Complete
+    Insert
+
+    def self.from_s?(str_mode) : OpenAIMode?
+      return OpenAIMode.parse?(str_mode)
+      # return OpenAIMode.values.find do |openai_mode|
+      #  openai_mode.to_s.downcase == str_mode.downcase
+      # end
+    end
+  end
+
+  def self.context_token_limit(model : OpenAIModel) : UInt32
+    hash = {
+      OpenAIModel::Gpt_3_5_Turbo => 1024_u32 * 4_u32, 
+      OpenAIModel::Gpt_3_5_Turbo_16k => 1024_u32 * 16_u32, 
+      OpenAIModel::Gpt_4 => 1024_u32 * 8_u32,
+    }
+
+    return 0_u32 unless hash[model]?
+    return hash[model]
+end
+
+class GptConfig
+  # defaults
+  # property temperature = 0.82
+  property model : OpenAIModel = OpenAIModel::Gpt_3_5_Turbo
+  # property model = "gpt-3.5-turbo-16k"
+  # property prompt_size = 10_000
+  property prompt_size = 2_700
+  property temperature = 0.8
+  property presence_penalty = 1.0
+  property frequency_penalty = 1.0
+  property max_tokens = 384
+  property mode = OpenAIMode::Chat
+  property openai_key : String = ENV.fetch("OPENAI_API_KEY")
+end
+end
+

src/parsers/prompt_string.cr

@@ -1,5 +1,5 @@
 
-require "../prompt"
+require "../prompts/generic_prompt"
 
 module Parser
   class PromptString

src/requests/chat.cr

@@ -0,0 +1,7 @@
+
+require "./generic_request"
+
+module Requests
+  class ChatRequest < GenericRequest
+  end
+end

src/requests/complete.cr

@@ -0,0 +1,8 @@
+
+
+require "./generic_request"
+
+module Requests
+  class CompleteRequest < GenericRequest
+  end
+end

src/requests/generic_request.cr

@@ -0,0 +1,23 @@
+
+module Requests
+  class GenericRequest
+    getter context_size : Int32 = 0
+
+    def initialize(@config : AIUtils::GptConfig)
+    end
+
+    def perform(messages)
+      openai = OpenAI::Client.new(access_token: @config.openai_key)
+      result = openai.chat(
+        @config.model.to_s, 
+        messages,
+        {
+          "temperature" => @config.temperature,
+          "presence_penalty" => @config.presence_penalty,
+          "frequency_penalty" => @config.frequency_penalty,
+          "max_tokens" => @config.max_tokens,
+        }
+      )
+    end
+  end
+end

src/requests/insert.cr

@@ -0,0 +1,8 @@
+
+
+require "./generic_request"
+
+module Requests
+  class InsertRequest < GenericRequest
+  end
+end

src/storyteller.cr

@@ -1,35 +1,6 @@
-enum OpenAIMode
-  Chat
-  Complete
-  Insert
 
-  def self.from_s(str_mode)
+require "./config"
-    return OpenAIMode.values.find do |openai_mode|
+require "./requests/chat"
-      openai_mode.to_s.downcase == str_mode.downcase
-    end
-  end
-end
-
-class GptConfig
-  # defaults
-  property gpt_temperature = 0.82
-  property gpt_presence_penalty = 1
-  property gpt_frequency_penalty = 1
-  property gpt_max_tokens = 256
-  property gpt_mode = OpenAIMode::Chat
-end
-
-class AppConfig
-  property input_file = STDIN
-  property input_file_path = ""
-  property output_file = STDOUT
-  property output_file_path = ""
-  property past_characters_limit = 1000
-  property use_color = true
-  property make_request = true
-  property verbose = false
-  property gpt_config = GptConfig.new
-end
 
 class Storyteller
   property config = AppConfig.new
@@ -79,26 +50,35 @@ class Storyteller
 
       parser.separator("GPT options")
 
-      parser.on("--gpt-mode MODE", "GPT mode (chat,insert,complete) (default: chat)") do |chosen_mode|
+      parser.on("--gpt-model MODEL", "GPT model (...) (default: gpt-3.5-turbo)") do |chosen_model|
-        result_mode = OpenAIMode.from_s(chosen_mode.downcase)
+        result_model = AIUtils::OpenAIModel.from_s?(chosen_model.downcase)
-        if result_mode.nil?
+        if result_model.nil?
-          STDERR.puts "ERROR: unknown mode #{chosen_mode}"
+          STDERR.puts "ERROR: unknown model #{chosen_model}"
           exit 1
         end
-        gpt_mode = result_mode unless result_mode.nil?
+        @config.gpt_config.model = result_model unless result_model.nil?
       end
 
-      parser.on("--gpt-temperature TEMPERATURE", "GPT Temperature") do |temperature|
+      # parser.on("--gpt-mode MODE", "GPT mode (chat,insert,complete) (default: chat)") do |chosen_mode|
-        gpt_temperature = temperature
+      #   result_mode = AIUtils::OpenAIMode.from_s(chosen_mode.downcase)
+      #   if result_mode.nil?
+      #     STDERR.puts "ERROR: unknown mode #{chosen_mode}"
+      #     exit 1
+      #   end
+      #   @config.gpt_config.mode = result_mode unless result_mode.nil?
+      # end
+
+      parser.on("--gpt-temperature TEMPERATURE", "GPT Temperature (default #{@config.gpt_config.temperature})") do |temperature|
+        @config.gpt_config.temperature = temperature.to_f
       end
-      parser.on("--gpt-presence-penalty PENALTY", "GPT Presence Penalty") do |presence_penalty|
+      parser.on("--gpt-presence-penalty PENALTY", "GPT Presence Penalty (default #{@config.gpt_config.presence_penalty})") do |presence_penalty|
-        gpt_presence_penalty = presence_penalty
+        @config.gpt_config.presence_penalty = presence_penalty.to_f
       end
-      parser.on("--gpt-frequency-penalty PENALTY", "GPT Frequency Penalty") do |frequency_penalty|
+      parser.on("--gpt-frequency-penalty PENALTY", "GPT Frequency Penalty (default #{@config.gpt_config.frequency_penalty})") do |frequency_penalty|
-        gpt_frequency_penalty = frequency_penalty
+        @config.gpt_config.frequency_penalty = frequency_penalty.to_f
       end
-      parser.on("--gpt-max-tokens TOKENS", "GPT Max Tokens") do |max_tokens|
+      parser.on("--gpt-max-tokens TOKENS", "GPT Max Tokens (default #{@config.gpt_config.max_tokens})") do |max_tokens|
-        gpt_max_tokens = max_tokens
+        @config.gpt_config.max_tokens = max_tokens.to_i
       end
 
     end
@@ -130,21 +110,15 @@ class Storyteller
     end
     self.write_file(@config.output_file, prompt, @config.use_color)
     @config.output_file.close
-
-    # rescue ex : OptionParser::InvalidOption
-    #   STDERR.puts @parser
-    #   STDERR.puts "\nERROR: #{ex.message}"
-    #   exit 1
-
-    # rescue ex: OpenAI::Client::ClientError
-    #   STDERR.puts "ERROR: #{ex.message}"
-    #   exit 1
   end
 
   def complete(prompt : Prompt, make_request : Bool, verbose : Bool)
     builder = Builder::OpenAIChat.new(verbose: verbose)
-    messages = builder.build(prompt)
+    context_token_limit = AIUtils.context_token_limit(@config.gpt_config.model)
+    messages = builder.build(prompt, context_token_limit)
 
+    STDERR.puts "model = #{@config.gpt_config.model}"
+    STDERR.puts "context_token_limit = #{context_token_limit}"
     STDERR.puts messages if verbose
 
     return prompt if !make_request
@@ -158,7 +132,7 @@ class Storyteller
       loop do
         tick += 1
         print "."
-        if tick > (3 * 60)
+        if tick > (2 * 60)
           print "(timeout)"
           exit 1
         end
@@ -167,20 +141,31 @@ class Storyteller
       end
     end
 
+    # TODO:
+    # request = Request::Chat.new(config: @config.gpt_config)
+    # result = request.perform(message)
+
     spawn do
-      openai = OpenAI::Client.new(access_token: ENV.fetch("OPENAI_API_KEY"))
+      request = Requests::ChatRequest.new(config: @config.gpt_config)
-      result = openai.chat(
+      result = request.perform(messages)
-        "gpt-3.5-turbo", 
+
-        messages,
+      pp result.choices if @config.verbose
-        {
-        "temperature" => 0.82,
-        "presence_penalty" => 1,
-        "frequency_penalty" => 1,
-        "max_tokens" => 256
-      }
-      )
       prompt.present_zone.content << result.choices.first["message"]["content"]
       channel_ready.send(true)
+    rescue ex: KeyError
+      puts "(openai error)"
+      STDERR.puts "ERROR: #{ex.message}"
+      exit 1
+
+    rescue ex: IO::Error
+      puts "(network error)"
+      STDERR.puts "ERROR: #{ex.message}"
+      exit 1
+
+    rescue ex: Socket::Addrinfo::Error
+      puts "(network error)"
+      STDERR.puts "ERROR: #{ex.message}"
+      exit 1
     end
 
     # sp.start
@@ -203,7 +188,7 @@ class Storyteller
 
   def write_file(output_file : IO::FileDescriptor, prompt : Prompt, use_color : Bool)
     # STDERR.puts "d: building builder"
-    builder = Builder::PromptString.new(use_color)
+    builder = ContextBuilder::PromptString.new(use_color)
     # STDERR.puts "d: building"
     text = builder.build(prompt)
     output_file.write_string(text.to_slice)