ruby_llm 0.1.0.pre26 → 0.1.0.pre28

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ea5a792c0bc3361e9784fc902e8e0e08f07cb2a6db03b408ee6362deb50c6edc
-  data.tar.gz: 44679aa95838fb8cbc8ff0ae42db0a5d2c9862d7af0ec461ea98c41276276bdf
+  metadata.gz: 467d0db634f70567417e20d87791372e39276e18fa4cd145cc065925b7379e7f
+  data.tar.gz: 70f3e92280366104d53446891ae645183706df5d8d384799c83299a43142e2eb
 SHA512:
-  metadata.gz: 3fa7cdaa35b5bfe2f60a25edfaefec82f89cf47d54fe8bad2f225fb4d339326fc341194fd734310e224f07055d9c0dccad81c6e610c6bd071aa64883c7b49a1d
-  data.tar.gz: aba94a44af3cb6348cb7bac016ad30a539ef6f980e16261dce324e9235214d869e5bf9ae78d46a7dc01ca1a7cfd0ec682a97b18872c1b50a6ed984bdd699b077
+  metadata.gz: 06e6021199954eace1c88e984f052d1d2644a9df3905280f9903fbd7ecfdf3b3bf904d3078c34982aeed43e15a1871946e1a9ad174ba621c7dcf3ce64ab190b5
+  data.tar.gz: 6697cd297ed9805114c433bf7610d4d608dce919b5ee3d99d1cec135d329640d607f64d3282ace206cb176f30390defc9c14ced6842f659033d96c04e7853dbe

data/.rubocop.yml

@@ -1,6 +1,5 @@
 require:
   - rubocop-rake
-  - rubocop-rspec
 
 AllCops:
   TargetRubyVersion: 3.1

data/README.md

@@ -1,6 +1,6 @@
 # RubyLLM
 
-A delightful Ruby interface to the latest large language models. Stop wrestling with multiple APIs and inconsistent interfaces. RubyLLM gives you a clean, unified way to work with models from OpenAI, Anthropic, Google, and DeepSeek.
+A delightful Ruby way to work with AI language models. Provides a unified interface to OpenAI, Anthropic, Google, and DeepSeek models with automatic token counting, proper streaming support, and a focus on developer happiness. No wrapping your head around multiple APIs - just clean Ruby code that works.
 
 <p align="center">
   <img src="https://upload.wikimedia.org/wikipedia/commons/4/4d/OpenAI_Logo.svg" alt="OpenAI" height="40" width="120">
@@ -73,7 +73,7 @@ image_models = RubyLLM.models.image_models
 Conversations are simple and natural:
 
 ```ruby
-chat = RubyLLM.chat model: 'claude-3-opus-20240229'
+chat = RubyLLM.chat model: 'gemini-2.0-flash'
 
 # Ask questions
 response = chat.ask "What's your favorite Ruby feature?"
@@ -101,7 +101,7 @@ Need vector embeddings for your text? RubyLLM makes it simple:
 RubyLLM.embed "Hello, world!"
 
 # Use a specific model
-RubyLLM.embed "Ruby is awesome!", model: "text-embedding-3-large"
+RubyLLM.embed "Ruby is awesome!", model: "text-embedding-004"
 
 # Process multiple texts at once
 RubyLLM.embed([
@@ -165,7 +165,7 @@ search = Search.new repo: Document
 chat.with_tools search, Calculator
 
 # Configure as needed
-chat.with_model('claude-3-opus-20240229')
+chat.with_model('claude-3-5-sonnet-20241022')
     .with_temperature(0.9)
 
 chat.ask "What's 2+2?"
@@ -373,7 +373,7 @@ class WeatherTool < RubyLLM::Tool
 end
 
 # Use tools with your persisted chats
-chat = Chat.create! model_id: "gpt-4"
+chat = Chat.create! model_id: "deepseek-reasoner"
 chat.chat.with_tool WeatherTool.new
 
 # Ask about weather - tool usage is automatically saved

data/lib/ruby_llm/configuration.rb

@@ -19,7 +19,7 @@ module RubyLLM
                   :request_timeout
 
     def initialize
-      @request_timeout = 30
+      @request_timeout = 120
       @default_model = 'gpt-4o-mini'
       @default_embedding_model = 'text-embedding-3-small'
     end

data/lib/ruby_llm/model_capabilities/deepseek.rb

@@ -6,6 +6,9 @@ module RubyLLM
     module DeepSeek
       module_function
 
+      # Returns the context window size for the given model
+      # @param model_id [String] the model identifier
+      # @return [Integer] the context window size in tokens
       def context_window_for(model_id)
         case model_id
         when /deepseek-(?:chat|reasoner)/ then 64_000
@@ -13,34 +16,61 @@ module RubyLLM
         end
       end
 
-      def max_tokens_for(_model_id)
-        8_192
+      # Returns the maximum output tokens for the given model
+      # @param model_id [String] the model identifier
+      # @return [Integer] the maximum output tokens
+      def max_tokens_for(model_id)
+        case model_id
+        when /deepseek-(?:chat|reasoner)/ then 8_192
+        else 4_096 # Default if max_tokens not specified
+        end
       end
 
+      # Returns the input price per million tokens for the given model
+      # @param model_id [String] the model identifier
+      # @return [Float] the price per million tokens
       def input_price_for(model_id)
         PRICES.dig(model_family(model_id), :input_miss) || default_input_price
       end
 
+      # Returns the output price per million tokens for the given model
+      # @param model_id [String] the model identifier
+      # @return [Float] the price per million tokens
       def output_price_for(model_id)
         PRICES.dig(model_family(model_id), :output) || default_output_price
       end
 
+      # Returns the cache hit price per million tokens for the given model
+      # @param model_id [String] the model identifier
+      # @return [Float] the price per million tokens
       def cache_hit_price_for(model_id)
         PRICES.dig(model_family(model_id), :input_hit) || default_cache_hit_price
       end
 
+      # Determines if the model supports vision capabilities
+      # @param model_id [String] the model identifier
+      # @return [Boolean] true if the model supports vision
       def supports_vision?(_model_id)
-        true # Both deepseek-chat and deepseek-reasoner support vision
+        false # DeepSeek models don't currently support vision
       end
 
-      def supports_functions?(_model_id)
-        true # Both models support function calling
+      # Determines if the model supports function calling
+      # @param model_id [String] the model identifier
+      # @return [Boolean] true if the model supports function calling
+      def supports_functions?(model_id)
+        model_id.match?(/deepseek-chat/) # Only deepseek-chat supports function calling
       end
 
-      def supports_json_mode?(_model_id)
-        true # Both models support JSON mode
+      # Determines if the model supports JSON mode
+      # @param model_id [String] the model identifier
+      # @return [Boolean] true if the model supports JSON mode
+      def supports_json_mode?(model_id)
+        model_id.match?(/deepseek-chat/) # Only deepseek-chat supports JSON mode
       end
 
+      # Formats the model ID into a display name
+      # @param model_id [String] the model identifier
+      # @return [String] the formatted display name
       def format_display_name(model_id)
         case model_id
         when 'deepseek-chat' then 'DeepSeek V3'
@@ -52,18 +82,25 @@ module RubyLLM
         end
       end
 
+      # Returns the model type
+      # @param model_id [String] the model identifier
+      # @return [String] the model type
       def model_type(_model_id)
-        'chat' # Both models are chat models
+        'chat' # All DeepSeek models are chat models
       end
 
+      # Returns the model family for pricing purposes
+      # @param model_id [String] the model identifier
+      # @return [String] the model family identifier
       def model_family(model_id)
         case model_id
-        when /deepseek-chat/ then 'deepseek'
-        when /deepseek-reasoner/ then 'deepseek_reasoner'
-        else 'deepseek' # Default to base deepseek family
+        when /deepseek-chat/ then :chat
+        when /deepseek-reasoner/ then :reasoner
+        else :chat # Default to chat family
         end
       end
 
+      # Pricing information for DeepSeek models (USD per 1M tokens)
       PRICES = {
         chat: {
           input_hit: 0.07,   # $0.07 per million tokens on cache hit
@@ -77,6 +114,8 @@ module RubyLLM
         }
       }.freeze
 
+      private
+
       def default_input_price
         0.27 # Default to chat cache miss price
       end

data/lib/ruby_llm/model_capabilities/gemini.rb

@@ -4,14 +4,14 @@ module RubyLLM
   module ModelCapabilities
     # Determines capabilities and pricing for Google Gemini models
     module Gemini # rubocop:disable Metrics/ModuleLength
-      module_function
+      extend self
 
       def context_window_for(model_id)
         case model_id
-        when /gemini-2\.0-flash/ then 1_048_576
+        when /gemini-2\.0-flash/, /gemini-1\.5-flash/ then 1_048_576
         when /gemini-1\.5-pro/ then 2_097_152
-        when /gemini-1\.5/ then 1_048_576
         when /text-embedding/, /embedding-001/ then 2_048
+        when /aqa/ then 7_168
         else 32_768 # Sensible default for unknown models
         end
       end
@@ -26,45 +26,45 @@ module RubyLLM
       end
 
       def input_price_for(model_id)
-        PRICES.dig(pricing_family(model_id), :input) || default_input_price
+        base_price = PRICES.dig(pricing_family(model_id), :input) || default_input_price
+        return base_price unless long_context_model?(model_id)
+
+        # Double the price for prompts longer than 128k tokens
+        context_length(model_id) > 128_000 ? base_price * 2 : base_price
       end
 
       def output_price_for(model_id)
-        PRICES.dig(pricing_family(model_id), :output) || default_output_price
+        base_price = PRICES.dig(pricing_family(model_id), :output) || default_output_price
+        return base_price unless long_context_model?(model_id)
+
+        # Double the price for prompts longer than 128k tokens
+        context_length(model_id) > 128_000 ? base_price * 2 : base_price
       end
 
       def supports_vision?(model_id)
         return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-        return false if model_id.match?(/flash-lite/)
-        return false if model_id.match?(/imagen/)
+        return false if model_id.match?(/gemini-1\.0/)
 
-        # Only pro and regular flash models support vision
-        model_id.match?(/gemini-[12]\.(?:5|0)-(?:pro|flash)(?!-lite)/)
+        model_id.match?(/gemini-[12]\.[05]/)
       end
 
       def supports_functions?(model_id)
         return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-        return false if model_id.match?(/imagen/)
         return false if model_id.match?(/flash-lite/)
-        return false if model_id.match?(/bison|gecko|evergreen/)
+        return false if model_id.match?(/gemini-1\.0/)
 
-        # Currently only full models support function calling
-        model_id.match?(/gemini-[12]\.(?:5|0)-(?:pro|flash)(?!-lite)/)
+        model_id.match?(/gemini-[12]\.[05]-(?:pro|flash)(?!-lite)/)
       end
 
       def supports_json_mode?(model_id)
         return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-        return false if model_id.match?(/imagen/)
         return false if model_id.match?(/flash-lite/)
-        return false if model_id.match?(/bison|gecko|evergreen/)
+        return false if model_id.match?(/gemini-1\.0/)
 
-        # Gemini 1.5+ models support JSON mode
-        model_id.match?(/gemini-[12]\.(?:5|0)-(?:pro|flash)(?!-lite)/)
+        model_id.match?(/gemini-1\.5/)
       end
 
       def format_display_name(model_id)
-        return model_id unless model_id.start_with?('models/')
-
         model_id
           .delete_prefix('models/')
           .split('-')
@@ -72,14 +72,28 @@ module RubyLLM
           .join(' ')
           .gsub(/(\d+\.\d+)/, ' \1') # Add space before version numbers
           .gsub(/\s+/, ' ')          # Clean up multiple spaces
+          .gsub(/Aqa/, 'AQA')        # Special case for AQA
           .strip
       end
 
+      def supports_caching?(model_id)
+        return false if model_id.match?(/flash-lite|gemini-1\.0/)
+
+        model_id.match?(/gemini-[12]\.[05]/)
+      end
+
+      def supports_tuning?(model_id)
+        model_id.match?(/gemini-1\.5-flash/)
+      end
+
+      def supports_audio?(model_id)
+        model_id.match?(/gemini-[12]\.[05]/)
+      end
+
       def model_type(model_id)
         case model_id
         when /text-embedding|embedding/ then 'embedding'
         when /imagen/ then 'image'
-        when /bison|text-bison/ then 'legacy'
         else 'chat'
         end
       end
@@ -91,33 +105,77 @@ module RubyLLM
         when /gemini-1\.5-flash-8b/ then 'gemini15_flash_8b'
         when /gemini-1\.5-flash/ then 'gemini15_flash'
         when /gemini-1\.5-pro/ then 'gemini15_pro'
+        when /gemini-1\.0-pro/ then 'gemini10_pro'
         when /text-embedding-004/ then 'embedding4'
         when /embedding-001/ then 'embedding1'
-        when /bison|text-bison/ then 'bison'
-        when /imagen/ then 'imagen3'
+        when /aqa/ then 'aqa'
         else 'other'
         end
       end
 
-      def pricing_family(model_id)
+      def pricing_family(model_id) # rubocop:disable Metrics/CyclomaticComplexity
         case model_id
         when /gemini-2\.0-flash-lite/ then :flash_lite_2 # rubocop:disable Naming/VariableNumber
         when /gemini-2\.0-flash/ then :flash_2 # rubocop:disable Naming/VariableNumber
         when /gemini-1\.5-flash-8b/ then :flash_8b
         when /gemini-1\.5-flash/ then :flash
         when /gemini-1\.5-pro/ then :pro
+        when /gemini-1\.0-pro/ then :pro_1_0 # rubocop:disable Naming/VariableNumber
         when /text-embedding|embedding/ then :embedding
         else :base
         end
       end
 
+      private
+
+      def long_context_model?(model_id)
+        model_id.match?(/gemini-1\.5-(?:pro|flash)/)
+      end
+
+      def context_length(model_id)
+        context_window_for(model_id)
+      end
+
       PRICES = {
-        flash_2: { input: 0.10, output: 0.40 },       # Gemini 2.0 Flash # rubocop:disable Naming/VariableNumber
-        flash_lite_2: { input: 0.075, output: 0.30 }, # Gemini 2.0 Flash Lite # rubocop:disable Naming/VariableNumber
-        flash: { input: 0.075, output: 0.30 },        # Gemini 1.5 Flash basic pricing
-        flash_8b: { input: 0.0375, output: 0.15 },    # Gemini 1.5 Flash 8B
-        pro: { input: 1.25, output: 5.0 },            # Gemini 1.5 Pro
-        embedding: { input: 0.00, output: 0.00 }      # Text Embedding models are free
+        flash_2: { # Gemini 2.0 Flash # rubocop:disable Naming/VariableNumber
+          input: 0.10,
+          output: 0.40,
+          audio_input: 0.70,
+          cache: 0.025,
+          cache_storage: 1.00
+        },
+        flash_lite_2: { # Gemini 2.0 Flash Lite # rubocop:disable Naming/VariableNumber
+          input: 0.075,
+          output: 0.30,
+          cache: 0.01875,
+          cache_storage: 1.00
+        },
+        flash: { # Gemini 1.5 Flash
+          input: 0.075,
+          output: 0.30,
+          cache: 0.01875,
+          cache_storage: 1.00
+        },
+        flash_8b: { # Gemini 1.5 Flash 8B
+          input: 0.0375,
+          output: 0.15,
+          cache: 0.01,
+          cache_storage: 0.25
+        },
+        pro: { # Gemini 1.5 Pro
+          input: 1.25,
+          output: 5.0,
+          cache: 0.3125,
+          cache_storage: 4.50
+        },
+        pro_1_0: { # Gemini 1.0 Pro # rubocop:disable Naming/VariableNumber
+          input: 0.50,
+          output: 1.50
+        },
+        embedding: { # Text Embedding models
+          input: 0.00,
+          output: 0.00
+        }
       }.freeze
 
       def default_input_price

data/lib/ruby_llm/model_capabilities/openai.rb

@@ -8,23 +8,26 @@ module RubyLLM
 
       def context_window_for(model_id)
         case model_id
-        when /gpt-4o/, /o1/, /gpt-4-turbo/ then 128_000
-        when /gpt-4-0[0-9]{3}/            then 8_192
-        when /gpt-3.5-turbo-instruct/     then 4_096
-        when /gpt-3.5/                    then 16_385
-        else                                   4_096
+        when /o[13]-mini/, /o3-mini-2025/         then 200_000
+        when /o1-2024/                            then 200_000
+        when /gpt-4o/, /gpt-4-turbo/             then 128_000
+        when /gpt-4-0[0-9]{3}/                   then 8_192
+        when /gpt-3.5-turbo-instruct/            then 4_096
+        when /gpt-3.5/                           then 16_385
+        else 4_096
         end
       end
 
-      def max_tokens_for(model_id)
+      def max_tokens_for(model_id) # rubocop:disable Metrics/CyclomaticComplexity
         case model_id
-        when /o1-2024-12-17/              then 100_000
-        when /o1-mini-2024-09-12/         then 65_536
-        when /o1-preview-2024-09-12/      then 32_768
-        when /gpt-4o/, /gpt-4-turbo/      then 16_384
-        when /gpt-4-0[0-9]{3}/           then 8_192
-        when /gpt-3.5-turbo/             then 4_096
-        else                                  4_096
+        when /o1-2024/, /o3-mini/                then 100_000
+        when /o1-mini-2024/                      then 65_536
+        when /gpt-4o-2024-05-13/                then 4_096
+        when /gpt-4o/, /gpt-4o-mini/            then 16_384
+        when /gpt-4o-realtime/                  then 4_096
+        when /gpt-4-0[0-9]{3}/                  then 8_192
+        when /gpt-3.5-turbo/                    then 4_096
+        else 4_096
         end
       end
 
@@ -37,13 +40,17 @@ module RubyLLM
       end
 
       def supports_vision?(model_id)
-        model_id.include?('vision') || model_id.match?(/gpt-4-(?!0314|0613)/)
+        model_id.match?(/gpt-4o|o1/) || model_id.match?(/gpt-4-(?!0314|0613)/)
       end
 
       def supports_functions?(model_id)
         !model_id.include?('instruct')
       end
 
+      def supports_audio?(model_id)
+        model_id.match?(/audio-preview|realtime-preview|whisper|tts/)
+      end
+
       def supports_json_mode?(model_id)
         model_id.match?(/gpt-4-\d{4}-preview/) ||
           model_id.include?('turbo') ||
@@ -65,17 +72,24 @@ module RubyLLM
         end
       end
 
+      def supports_structured_output?(model_id)
+        model_id.match?(/gpt-4o|o[13]-mini|o1/)
+      end
+
       def model_family(model_id) # rubocop:disable Metrics/AbcSize,Metrics/CyclomaticComplexity,Metrics/MethodLength
         case model_id
-        when /o1-2024|o1-mini-2024/ then 'o1'
+        when /o3-mini/ then 'o3_mini'
         when /o1-mini/ then 'o1_mini'
         when /o1/ then 'o1'
+        when /gpt-4o-audio/ then 'gpt4o_audio'
         when /gpt-4o-realtime/ then 'gpt4o_realtime'
+        when /gpt-4o-mini-audio/ then 'gpt4o_mini_audio'
         when /gpt-4o-mini-realtime/ then 'gpt4o_mini_realtime'
         when /gpt-4o-mini/ then 'gpt4o_mini'
         when /gpt-4o/ then 'gpt4o'
         when /gpt-4-turbo/ then 'gpt4_turbo'
         when /gpt-4/ then 'gpt4'
+        when /gpt-3.5-turbo-instruct/ then 'gpt35_instruct'
         when /gpt-3.5/ then 'gpt35'
         when /dall-e-3/ then 'dalle3'
         when /dall-e-2/ then 'dalle2'
@@ -95,14 +109,51 @@ module RubyLLM
       private
 
       PRICES = {
-        o1: { input: 15.0, output: 60.0 },
-        o1_mini: { input: 3.0, output: 12.0 },
-        gpt4o_realtime: { input: 5.0, output: 20.0 },
-        gpt4o_mini_realtime: { input: 0.60, output: 2.40 },
-        gpt4o_mini: { input: 0.15, output: 0.60 },
-        gpt4o: { input: 2.50, output: 10.0 },
+        o1: { input: 15.0, cached_input: 7.5, output: 60.0 },
+        o1_mini: { input: 1.10, cached_input: 0.55, output: 4.40 },
+        o3_mini: { input: 1.10, cached_input: 0.55, output: 4.40 },
+        gpt4o: { input: 2.50, cached_input: 1.25, output: 10.0 },
+        gpt4o_audio: {
+          text_input: 2.50,
+          audio_input: 40.0,
+          text_output: 10.0,
+          audio_output: 80.0
+        },
+        gpt4o_realtime: {
+          text_input: 5.0,
+          cached_text_input: 2.50,
+          audio_input: 40.0,
+          cached_audio_input: 2.50,
+          text_output: 20.0,
+          audio_output: 80.0
+        },
+        gpt4o_mini: { input: 0.15, cached_input: 0.075, output: 0.60 },
+        gpt4o_mini_audio: {
+          text_input: 0.15,
+          audio_input: 10.0,
+          text_output: 0.60,
+          audio_output: 20.0
+        },
+        gpt4o_mini_realtime: {
+          text_input: 0.60,
+          cached_text_input: 0.30,
+          audio_input: 10.0,
+          cached_audio_input: 0.30,
+          text_output: 2.40,
+          audio_output: 20.0
+        },
         gpt4_turbo: { input: 10.0, output: 30.0 },
-        gpt35: { input: 0.50, output: 1.50 }
+        gpt4: { input: 30.0, output: 60.0 },
+        gpt35: { input: 0.50, output: 1.50 },
+        gpt35_instruct: { input: 1.50, output: 2.0 },
+        embedding3_large: { price: 0.13 },
+        embedding3_small: { price: 0.02 },
+        embedding2: { price: 0.10 },
+        davinci: { input: 2.0, output: 2.0 },
+        babbage: { input: 0.40, output: 0.40 },
+        tts1: { price: 15.0 },
+        tts1_hd: { price: 30.0 },
+        whisper1: { price: 0.006 }
       }.freeze
 
       def default_input_price
@@ -124,13 +175,13 @@ module RubyLLM
         name
           .gsub(/(\d{4}) (\d{2}) (\d{2})/, '\1\2\3')
           .gsub(/^Gpt /, 'GPT-')
-          .gsub(/^O1 /, 'O1-')
+          .gsub(/^O[13] /, 'O\1-')
           .gsub(/^Chatgpt /, 'ChatGPT-')
           .gsub(/^Tts /, 'TTS-')
           .gsub(/^Dall E /, 'DALL-E-')
           .gsub(/3\.5 /, '3.5-')
           .gsub(/4 /, '4-')
-          .gsub(/4o (?=Mini|Preview|Turbo)/, '4o-')
+          .gsub(/4o (?=Mini|Preview|Turbo|Audio)/, '4o-')
           .gsub(/\bHd\b/, 'HD')
       end
     end