ruby_llm 1.6.2 → 1.6.3

data/lib/ruby_llm/providers/gemini/capabilities.rb

@@ -7,9 +7,6 @@ module RubyLLM
       module Capabilities
         module_function
 
-        # Returns the context window size (input token limit) 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 /gemini-2\.5-pro-exp-03-25/, /gemini-2\.0-flash/, /gemini-2\.0-flash-lite/, /gemini-1\.5-flash/, /gemini-1\.5-flash-8b/ # rubocop:disable Layout/LineLength
@@ -18,70 +15,49 @@ module RubyLLM
           when /gemini-embedding-exp/ then 8_192
           when /text-embedding-004/, /embedding-001/ then 2_048
           when /aqa/ then 7_168
-          when /imagen-3/ then nil # No token limit for image generation
-          else 32_768 # Sensible default for unknown models
+          when /imagen-3/ then nil
+          else 32_768
           end
         end
 
-        # 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 /gemini-2\.5-pro-exp-03-25/ then 64_000
           when /gemini-2\.0-flash/, /gemini-2\.0-flash-lite/, /gemini-1\.5-flash/, /gemini-1\.5-flash-8b/, /gemini-1\.5-pro/ # rubocop:disable Layout/LineLength
             8_192
-          when /gemini-embedding-exp/ then nil # Elastic, supports 3072, 1536, or 768
-          when /text-embedding-004/, /embedding-001/ then 768 # Output dimension size for embeddings
-          when /aqa/ then 1_024
-          when /imagen-3/ then 4 # Output images
-          else 4_096 # Sensible default
+          when /gemini-embedding-exp/ then nil
+          when /text-embedding-004/, /embedding-001/ then 768
+          when /imagen-3/ then 4
+          else 4_096
           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 in USD
         def input_price_for(model_id)
           base_price = PRICES.dig(pricing_family(model_id), :input) || default_input_price
           return base_price unless long_context_model?(model_id)
 
-          # Apply different pricing for prompts longer than 128k tokens
           context_window_for(model_id) > 128_000 ? base_price * 2 : base_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 in USD
         def output_price_for(model_id)
           base_price = PRICES.dig(pricing_family(model_id), :output) || default_output_price
           return base_price unless long_context_model?(model_id)
 
-          # Apply different pricing for prompts longer than 128k tokens
           context_window_for(model_id) > 128_000 ? base_price * 2 : base_price
         end
 
-        # Determines if the model supports vision (image/video) inputs
-        # @param model_id [String] the model identifier
-        # @return [Boolean] true if the model supports vision inputs
         def supports_vision?(model_id)
           return false if model_id.match?(/text-embedding|embedding-001|aqa/)
 
           model_id.match?(/gemini|flash|pro|imagen/)
         end
 
-        # 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)
           return false if model_id.match?(/text-embedding|embedding-001|aqa|flash-lite|imagen|gemini-2\.0-flash-lite/)
 
           model_id.match?(/gemini|pro|flash/)
         end
 
-        # 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)
           if model_id.match?(/text-embedding|embedding-001|aqa|imagen|gemini-2\.0-flash-lite|gemini-2\.5-pro-exp-03-25/)
             return false
@@ -90,24 +66,18 @@ module RubyLLM
           model_id.match?(/gemini|pro|flash/)
         end
 
-        # Formats the model ID into a human-readable display name
-        # @param model_id [String] the model identifier
-        # @return [String] the formatted display name
         def format_display_name(model_id)
           model_id
             .delete_prefix('models/')
             .split('-')
             .map(&:capitalize)
             .join(' ')
-            .gsub(/(\d+\.\d+)/, ' \1') # Add space before version numbers
-            .gsub(/\s+/, ' ')          # Clean up multiple spaces
-            .gsub('Aqa', 'AQA')        # Special case for AQA
+            .gsub(/(\d+\.\d+)/, ' \1')
+            .gsub(/\s+/, ' ')
+            .gsub('Aqa', 'AQA')
             .strip
         end
 
-        # Determines if the model supports context caching
-        # @param model_id [String] the model identifier
-        # @return [Boolean] true if the model supports caching
         def supports_caching?(model_id)
           if model_id.match?(/flash-lite|gemini-2\.5-pro-exp-03-25|aqa|imagen|text-embedding|embedding-001/)
             return false
@@ -116,23 +86,14 @@ module RubyLLM
           model_id.match?(/gemini|pro|flash/)
         end
 
-        # Determines if the model supports tuning
-        # @param model_id [String] the model identifier
-        # @return [Boolean] true if the model supports tuning
         def supports_tuning?(model_id)
           model_id.match?(/gemini-1\.5-flash|gemini-1\.5-flash-8b/)
         end
 
-        # Determines if the model supports audio inputs
-        # @param model_id [String] the model identifier
-        # @return [Boolean] true if the model supports audio inputs
         def supports_audio?(model_id)
           model_id.match?(/gemini|pro|flash/)
         end
 
-        # Returns the type of model (chat, embedding, image)
-        # @param model_id [String] the model identifier
-        # @return [String] the model type
         def model_type(model_id)
           case model_id
           when /text-embedding|embedding|gemini-embedding/ then 'embedding'
@@ -141,9 +102,6 @@ module RubyLLM
           end
         end
 
-        # Returns the model family identifier
-        # @param model_id [String] the model identifier
-        # @return [String] the model family identifier
         def model_family(model_id)
           case model_id
           when /gemini-2\.5-pro-exp-03-25/ then 'gemini25_pro_exp'
@@ -161,9 +119,6 @@ module RubyLLM
           end
         end
 
-        # Returns the pricing family identifier for the model
-        # @param model_id [String] the model identifier
-        # @return [Symbol] the pricing family identifier
         def pricing_family(model_id)
           case model_id
           when /gemini-2\.5-pro-exp-03-25/ then :pro_2_5 # rubocop:disable Naming/VariableNumber
@@ -180,86 +135,75 @@ module RubyLLM
           end
         end
 
-        # Determines if the model supports long context
-        # @param model_id [String] the model identifier
-        # @return [Boolean] true if the model supports long context
         def long_context_model?(model_id)
           model_id.match?(/gemini-1\.5-(?:pro|flash)|gemini-1\.5-flash-8b/)
         end
 
-        # Returns the context length for the model
-        # @param model_id [String] the model identifier
-        # @return [Integer] the context length in tokens
         def context_length(model_id)
           context_window_for(model_id)
         end
 
-        # Pricing information for Gemini models (per 1M tokens in USD)
         PRICES = {
-          flash_2: { # Gemini 2.0 Flash # rubocop:disable Naming/VariableNumber
+          flash_2: { # rubocop:disable Naming/VariableNumber
             input: 0.10,
             output: 0.40,
             audio_input: 0.70,
             cache: 0.025,
             cache_storage: 1.00,
-            grounding_search: 35.00 # per 1K requests after 1.5K free
+            grounding_search: 35.00
           },
-          flash_lite_2: { # Gemini 2.0 Flash Lite # rubocop:disable Naming/VariableNumber
+          flash_lite_2: { # rubocop:disable Naming/VariableNumber
             input: 0.075,
             output: 0.30
           },
-          flash: { # Gemini 1.5 Flash
+          flash: {
             input: 0.075,
             output: 0.30,
             cache: 0.01875,
             cache_storage: 1.00,
-            grounding_search: 35.00 # per 1K requests up to 5K per day
+            grounding_search: 35.00
           },
-          flash_8b: { # Gemini 1.5 Flash 8B
+          flash_8b: {
             input: 0.0375,
             output: 0.15,
             cache: 0.01,
             cache_storage: 0.25,
-            grounding_search: 35.00 # per 1K requests up to 5K per day
+            grounding_search: 35.00
           },
-          pro: { # Gemini 1.5 Pro
+          pro: {
             input: 1.25,
             output: 5.0,
             cache: 0.3125,
             cache_storage: 4.50,
-            grounding_search: 35.00 # per 1K requests up to 5K per day
+            grounding_search: 35.00
           },
-          pro_2_5: { # Gemini 2.5 Pro Experimental # rubocop:disable Naming/VariableNumber
+          pro_2_5: { # rubocop:disable Naming/VariableNumber
             input: 0.12,
             output: 0.50
           },
-          gemini_embedding: { # Gemini Embedding Experimental
+          gemini_embedding: {
             input: 0.002,
             output: 0.004
           },
-          embedding: { # Text Embedding models
+          embedding: {
             input: 0.00,
             output: 0.00
           },
-          imagen: { # Imagen 3
-            price: 0.03 # per image
+          imagen: {
+            price: 0.03
           },
-          aqa: { # AQA model
+          aqa: {
             input: 0.00,
             output: 0.00
           }
         }.freeze
 
-        # Default input price for unknown models
-        # @return [Float] the default input price per million tokens
         def default_input_price
-          0.075 # Default to Flash pricing
+          0.075
         end
 
-        # Default output price for unknown models
-        # @return [Float] the default output price per million tokens
         def default_output_price
-          0.30 # Default to Flash pricing
+          0.30
         end
 
         def modalities_for(model_id)
@@ -268,19 +212,13 @@ module RubyLLM
             output: ['text']
           }
 
-          # Vision support
           if supports_vision?(model_id)
             modalities[:input] << 'image'
             modalities[:input] << 'pdf'
           end
 
-          # Audio support
           modalities[:input] << 'audio' if model_id.match?(/audio/)
-
-          # Embedding output
           modalities[:output] << 'embeddings' if model_id.match?(/embedding|gemini-embedding/)
-
-          # Image output for imagen models
           modalities[:output] = ['image'] if model_id.match?(/imagen/)
 
           modalities
@@ -289,21 +227,11 @@ module RubyLLM
         def capabilities_for(model_id)
           capabilities = ['streaming']
 
-          # Function calling
           capabilities << 'function_calling' if supports_functions?(model_id)
-
-          # JSON mode
           capabilities << 'structured_output' if supports_json_mode?(model_id)
-
-          # Batch processing
           capabilities << 'batch' if model_id.match?(/embedding|flash/)
-
-          # Caching
           capabilities << 'caching' if supports_caching?(model_id)
-
-          # Tuning
           capabilities << 'fine_tuning' if supports_tuning?(model_id)
-
           capabilities
         end
 
@@ -316,10 +244,8 @@ module RubyLLM
             output_per_million: prices[:output]
           }
 
-          # Add cached pricing if available
           standard_pricing[:cached_input_per_million] = prices[:input_hit] if prices[:input_hit]
 
-          # Batch pricing (typically 50% discount)
           batch_pricing = {
             input_per_million: (standard_pricing[:input_per_million] || 0) * 0.5,
             output_per_million: (standard_pricing[:output_per_million] || 0) * 0.5
@@ -336,7 +262,6 @@ module RubyLLM
             }
           }
 
-          # Add embedding pricing if applicable
           if model_id.match?(/embedding|gemini-embedding/)
             pricing[:embeddings] = {
               standard: { input_per_million: prices[:price] || 0.002 }

data/lib/ruby_llm/providers/gemini/chat.rb

@@ -12,14 +12,14 @@ module RubyLLM
         end
 
         def render_payload(messages, tools:, temperature:, model:, stream: false, schema: nil) # rubocop:disable Metrics/ParameterLists,Lint/UnusedMethodArgument
-          @model = model # Store model for completion_url/stream_url
+          @model = model
           payload = {
             contents: format_messages(messages),
-            generationConfig: {
-              temperature: temperature
-            }
+            generationConfig: {}
           }
 
+          payload[:generationConfig][:temperature] = temperature unless temperature.nil?
+
           if schema
             payload[:generationConfig][:responseMimeType] = 'application/json'
             payload[:generationConfig][:responseSchema] = convert_schema_to_gemini(schema)
@@ -43,7 +43,7 @@ module RubyLLM
         def format_role(role)
           case role
           when :assistant then 'model'
-          when :system, :tool then 'user' # Gemini doesn't have system, use user role, function responses use user role
+          when :system, :tool then 'user'
           else role.to_s
           end
         end
@@ -118,10 +118,8 @@ module RubyLLM
           candidate = data.dig('candidates', 0)
           return '' unless candidate
 
-          # Content will be empty for function calls
           return '' if function_call?(candidate)
 
-          # Extract text content
           parts = candidate.dig('content', 'parts')
           text_parts = parts&.select { |p| p['text'] }
           return '' unless text_parts&.any?

data/lib/ruby_llm/providers/gemini/embeddings.rb

@@ -17,8 +17,6 @@ module RubyLLM
 
         def parse_embedding_response(response, model:, text:)
           vectors = response.body['embeddings']&.map { |e| e['values'] }
-          # If we only got one embedding AND the input was a single string (not an array),
-          # return it as a single vector
           vectors = vectors.first if vectors&.length == 1 && !text.is_a?(Array)
 
           Embedding.new(vectors:, model:, input_tokens: 0)

data/lib/ruby_llm/providers/gemini/images.rb

@@ -32,7 +32,6 @@ module RubyLLM
             raise Error, 'Unexpected response format from Gemini image generation API'
           end
 
-          # Extract mime type and base64 data
           mime_type = image_data['mimeType'] || 'image/png'
           base64_data = image_data['bytesBase64Encoded']
 

data/lib/ruby_llm/providers/gemini/media.rb

@@ -8,7 +8,6 @@ module RubyLLM
         module_function
 
         def format_content(content)
-          # Convert Hash/Array back to JSON string for API
           return [format_text(content.to_json)] if content.is_a?(Hash) || content.is_a?(Array)
           return [format_text(content)] unless content.is_a?(Content)
 

data/lib/ruby_llm/providers/gemini/models.rb

@@ -13,7 +13,6 @@ module RubyLLM
 
         def parse_list_models_response(response, slug, capabilities)
           Array(response.body['models']).map do |model_data|
-            # Extract model ID without "models/" prefix
             model_id = model_data['name'].gsub('models/', '')
 
             Model::Info.new(
@@ -21,7 +20,7 @@ module RubyLLM
               name: model_data['displayName'],
               provider: slug,
               family: capabilities.model_family(model_id),
-              created_at: nil, # Gemini API doesn't provide creation date
+              created_at: nil,
               context_window: model_data['inputTokenLimit'] || capabilities.context_window_for(model_id),
               max_output_tokens: model_data['outputTokenLimit'] || capabilities.max_tokens_for(model_id),
               modalities: capabilities.modalities_for(model_id),

data/lib/ruby_llm/providers/gemini/tools.rb

@@ -5,7 +5,6 @@ module RubyLLM
     class Gemini
       # Tools methods for the Gemini API implementation
       module Tools
-        # Format tools for Gemini API
         def format_tools(tools)
           return [] if tools.empty?
 
@@ -14,7 +13,6 @@ module RubyLLM
           }]
         end
 
-        # Extract tool calls from response data
         def extract_tool_calls(data)
           return nil unless data
 
@@ -43,7 +41,6 @@ module RubyLLM
 
         private
 
-        # Format a single tool for Gemini API
         def function_declaration_for(tool)
           {
             name: tool.name,
@@ -52,7 +49,6 @@ module RubyLLM
           }.compact
         end
 
-        # Format tool parameters for Gemini API
         def format_parameters(parameters)
           {
             type: 'OBJECT',
@@ -66,7 +62,6 @@ module RubyLLM
           }
         end
 
-        # Convert RubyLLM param types to Gemini API types
         def param_type_for_gemini(type)
           case type.to_s.downcase
           when 'integer', 'number', 'float' then 'NUMBER'

data/lib/ruby_llm/providers/gpustack/chat.rb

@@ -8,7 +8,6 @@ module RubyLLM
         module_function
 
         def format_role(role)
-          # GPUStack doesn't use the new OpenAI convention for system prompts
           role.to_s
         end
       end

data/lib/ruby_llm/providers/gpustack/models.rb

@@ -30,13 +30,12 @@ module RubyLLM
                 categories: model['categories']
               },
               context_window: model.dig('meta', 'n_ctx'),
-              # Using context window as max tokens since it's not explicitly provided
               max_tokens: model.dig('meta', 'n_ctx'),
               supports_vision: model.dig('meta', 'support_vision') || false,
               supports_functions: model.dig('meta', 'support_tool_calls') || false,
-              supports_json_mode: true, # Assuming all models support JSON mode
-              input_price_per_million: 0.0,  # Price information not available in new format
-              output_price_per_million: 0.0  # Price information not available in new format
+              supports_json_mode: true,
+              input_price_per_million: 0.0,
+              output_price_per_million: 0.0
             )
           end
         end

data/lib/ruby_llm/providers/mistral/capabilities.rb

@@ -8,22 +8,18 @@ module RubyLLM
         module_function
 
         def supports_streaming?(model_id)
-          # All chat models support streaming, but not embedding/moderation/OCR/transcription
           !model_id.match?(/embed|moderation|ocr|transcriptions/)
         end
 
         def supports_tools?(model_id)
-          # Most chat models support tools except embedding/moderation/OCR/voxtral/transcription
           !model_id.match?(/embed|moderation|ocr|voxtral|transcriptions|mistral-(tiny|small)-(2312|2402)/)
         end
 
         def supports_vision?(model_id)
-          # Models with vision capabilities
           model_id.match?(/pixtral|mistral-small-(2503|2506)|mistral-medium/)
         end
 
         def supports_json_mode?(model_id)
-          # Most chat models support JSON mode (structured output)
           !model_id.match?(/embed|moderation|ocr|voxtral|transcriptions/) && supports_tools?(model_id)
         end
 
@@ -58,11 +54,11 @@ module RubyLLM
         end
 
         def context_window_for(_model_id)
-          32_768 # Default for most Mistral models
+          32_768
         end
 
         def max_tokens_for(_model_id)
-          8192 # Default for most Mistral models
+          8192
         end
 
         def modalities_for(model_id)
@@ -97,7 +93,6 @@ module RubyLLM
             capabilities << 'structured_output' if supports_json_mode?(model_id)
             capabilities << 'vision' if supports_vision?(model_id)
 
-            # Model-specific capabilities
             capabilities << 'reasoning' if model_id.match?(/magistral/)
             capabilities << 'batch' unless model_id.match?(/voxtral|ocr|embed|moderation/)
             capabilities << 'fine_tuning' if model_id.match?(/mistral-(small|medium|large)|devstral/)
@@ -117,12 +112,10 @@ module RubyLLM
 
         def release_date_for(model_id)
           case model_id
-          # 2023 releases
           when 'open-mistral-7b', 'mistral-tiny' then '2023-09-27'
           when 'mistral-medium-2312', 'mistral-small-2312', 'mistral-small',
                'open-mixtral-8x7b', 'mistral-tiny-2312' then '2023-12-11'
 
-          # 2024 releases
           when 'mistral-embed' then '2024-01-11'
           when 'mistral-large-2402', 'mistral-small-2402' then '2024-02-26'
           when 'open-mixtral-8x22b', 'open-mixtral-8x22b-2404' then '2024-04-17'
@@ -140,7 +133,6 @@ module RubyLLM
           when 'codestral-2411-rc5', 'mistral-moderation-2411', 'mistral-moderation-latest' then '2024-11-26'
           when 'codestral-2412' then '2024-12-17'
 
-          # 2025 releases
           when 'mistral-small-2501' then '2025-01-13'
           when 'codestral-2501' then '2025-01-14'
           when 'mistral-saba-2502', 'mistral-saba-latest' then '2025-02-18'

data/lib/ruby_llm/providers/mistral/chat.rb

@@ -8,14 +8,12 @@ module RubyLLM
         module_function
 
         def format_role(role)
-          # Mistral doesn't use the new OpenAI convention for system prompts
           role.to_s
         end
 
         # rubocop:disable Metrics/ParameterLists
         def render_payload(messages, tools:, temperature:, model:, stream: false, schema: nil)
           payload = super
-          # Mistral doesn't support stream_options
           payload.delete(:stream_options)
           payload
         end

data/lib/ruby_llm/providers/mistral/embeddings.rb

@@ -12,7 +12,6 @@ module RubyLLM
         end
 
         def render_embedding_payload(text, model:, dimensions:) # rubocop:disable Lint/UnusedMethodArgument
-          # Mistral doesn't support dimensions parameter
           {
             model: model,
             input: text
@@ -24,8 +23,6 @@ module RubyLLM
           input_tokens = data.dig('usage', 'prompt_tokens') || 0
           vectors = data['data'].map { |d| d['embedding'] }
 
-          # If we only got one embedding AND the input was a single string (not an array),
-          # return it as a single vector
           vectors = vectors.first if vectors.length == 1 && !text.is_a?(Array)
 
           Embedding.new(vectors:, model:, input_tokens:)

data/lib/ruby_llm/providers/mistral/models.rb

@@ -21,7 +21,6 @@ module RubyLLM
           Array(response.body['data']).map do |model_data|
             model_id = model_data['id']
 
-            # Use fixed release date for Mistral models
             release_date = capabilities.release_date_for(model_id)
             created_at = release_date ? Time.parse(release_date) : nil
 

data/lib/ruby_llm/providers/ollama/chat.rb

@@ -19,7 +19,6 @@ module RubyLLM
         end
 
         def format_role(role)
-          # Ollama doesn't use the new OpenAI convention for system prompts
           role.to_s
         end
       end

data/lib/ruby_llm/providers/ollama/media.rb

@@ -10,7 +10,6 @@ module RubyLLM
         module_function
 
         def format_content(content)
-          # Convert Hash/Array back to JSON string for API
           return content.to_json if content.is_a?(Hash) || content.is_a?(Array)
           return content unless content.is_a?(Content)
 

data/lib/ruby_llm/providers/openai/capabilities.rb

@@ -235,20 +235,11 @@ module RubyLLM
 
           # Vision support
           modalities[:input] << 'image' if supports_vision?(model_id)
-
-          # Audio support
           modalities[:input] << 'audio' if model_id.match?(/whisper|audio|tts|transcribe/)
-
-          # PDF support
           modalities[:input] << 'pdf' if supports_vision?(model_id)
-
-          # Output modalities
           modalities[:output] << 'audio' if model_id.match?(/tts|audio/)
-
           modalities[:output] << 'image' if model_id.match?(/dall-e|image/)
-
           modalities[:output] << 'embeddings' if model_id.match?(/embedding/)
-
           modalities[:output] << 'moderation' if model_id.match?(/moderation/)
 
           modalities
@@ -257,13 +248,10 @@ module RubyLLM
         def capabilities_for(model_id) # rubocop:disable Metrics/PerceivedComplexity
           capabilities = []
 
-          # Common capabilities
           capabilities << 'streaming' unless model_id.match?(/moderation|embedding/)
           capabilities << 'function_calling' if supports_functions?(model_id)
           capabilities << 'structured_output' if supports_json_mode?(model_id)
           capabilities << 'batch' if model_id.match?(/embedding|batch/)
-
-          # Advanced capabilities
           capabilities << 'reasoning' if model_id.match?(/o\d|gpt-5|codex/)
 
           if model_id.match?(/gpt-4-turbo|gpt-4o/)
@@ -281,16 +269,13 @@ module RubyLLM
             output_per_million: output_price_for(model_id)
           }
 
-          # Add cached pricing if available
           if respond_to?(:cached_input_price_for)
             cached_price = cached_input_price_for(model_id)
             standard_pricing[:cached_input_per_million] = cached_price if cached_price
           end
 
-          # Pricing structure
           pricing = { text_tokens: { standard: standard_pricing } }
 
-          # Add batch pricing if applicable
           if model_id.match?(/embedding|batch/)
             pricing[:text_tokens][:batch] = {
               input_per_million: standard_pricing[:input_per_million] * 0.5,

data/lib/ruby_llm/providers/openai/chat.rb

@@ -18,13 +18,10 @@ module RubyLLM
             stream: stream
           }
 
-          # Only include temperature if it's not nil (some models don't accept it)
           payload[:temperature] = temperature unless temperature.nil?
-
           payload[:tools] = tools.map { |_, tool| tool_for(tool) } if tools.any?
 
           if schema
-            # Use strict mode from schema if specified, default to true
             strict = schema[:strict] != false
 
             payload[:response_format] = {