llm.rb 0.3.2 → 0.4.0

data/lib/llm/providers/anthropic/models.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+class LLM::Anthropic
+  ##
+  # The {LLM::Anthropic::Models LLM::Anthropic::Models} class provides a model
+  # object for interacting with [Anthropic's models API](https://platform.anthropic.com/docs/api-reference/models/list).
+  # The models API allows a client to query Anthropic for a list of models
+  # that are available for use with the Anthropic API.
+  #
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm = LLM.anthropic(ENV["KEY"])
+  #   res = llm.models.all
+  #   res.each do |model|
+  #     print "id: ", model.id, "\n"
+  #   end
+  class Models
+    ##
+    # Returns a new Models object
+    # @param provider [LLM::Provider]
+    # @return [LLM::Anthropic::Files]
+    def initialize(provider)
+      @provider = provider
+    end
+
+    ##
+    # List all models
+    # @example
+    #   llm = LLM.anthropic(ENV["KEY"])
+    #   res = llm.models.all
+    #   res.each do |model|
+    #     print "id: ", model.id, "\n"
+    #   end
+    # @see https://docs.anthropic.com/en/api/models-list Anthropic docs
+    # @param [Hash] params Other parameters (see Anthropic docs)
+    # @raise (see LLM::Provider#request)
+    # @return [LLM::Response::FileList]
+    def all(**params)
+      query = URI.encode_www_form(params)
+      req = Net::HTTP::Get.new("/v1/models?#{query}", headers)
+      res = request(http, req)
+      LLM::Response::ModelList.new(res).tap { |modellist|
+        models = modellist.body["data"].map do |model|
+          LLM::Model.from_hash(model).tap { _1.provider = @provider }
+        end
+        modellist.models = models
+      }
+    end
+
+    private
+
+    def http
+      @provider.instance_variable_get(:@http)
+    end
+
+    [:headers, :request].each do |m|
+      define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
+    end
+  end
+end

data/lib/llm/providers/anthropic.rb

@@ -8,6 +8,7 @@ module LLM
     require_relative "anthropic/error_handler"
     require_relative "anthropic/response_parser"
     require_relative "anthropic/format"
+    require_relative "anthropic/models"
     include Format
 
     HOST = "api.anthropic.com"
@@ -45,16 +46,27 @@ module LLM
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
+    # @raise [LLM::Error::PromptError]
+    #  When given an object a provider does not understand
     # @return (see LLM::Provider#complete)
-    def complete(prompt, role = :user, model: "claude-3-5-sonnet-20240620", max_tokens: 1024, **params)
-      params   = {max_tokens:, model:}.merge!(params)
-      req      = Net::HTTP::Post.new("/v1/messages", headers)
+    def complete(prompt, role = :user, model: default_model, max_tokens: 1024, **params)
+      params = {max_tokens:, model:}.merge!(params)
+      req = Net::HTTP::Post.new("/v1/messages", headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      req.body = JSON.dump({messages: format(messages)}.merge!(params))
-      res      = request(@http, req)
+      body = JSON.dump({messages: format(messages)}.merge!(params))
+      set_body_stream(req, StringIO.new(body))
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # Provides an interface to Anthropic's models API
+    # @see https://docs.anthropic.com/en/api/models-list
+    # @return [LLM::Anthropic::Models]
+    def models
+      LLM::Anthropic::Models.new(self)
+    end
+
     ##
     # @return (see LLM::Provider#assistant_role)
     def assistant_role
@@ -62,9 +74,11 @@ module LLM
     end
 
     ##
-    # @return (see LLM::Provider#models)
-    def models
-      @models ||= load_models!("anthropic")
+    # Returns the default model for chat completions
+    # @see https://docs.anthropic.com/en/docs/about-claude/models/all-models#model-comparison-table claude-3-5-sonnet-20240620
+    # @return [String]
+    def default_model
+      "claude-3-5-sonnet-20240620"
     end
 
     private

data/lib/llm/providers/gemini/format.rb

@@ -35,8 +35,13 @@ class LLM::Gemini
       when LLM::File
         file = content
         {inline_data: {mime_type: file.mime_type, data: file.to_b64}}
-      else
+      when String
         {text: content}
+      when LLM::Message
+        format_content(content.content)
+      else
+        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                       "is not supported by the Gemini API"
       end
     end
   end

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

@@ -13,7 +13,7 @@ class LLM::Gemini
   #
   #   llm = LLM.gemini(ENV["KEY"])
   #   res = llm.images.create prompt: "A dog on a rocket to the moon"
-  #   File.binwrite "rocket.png", res.images[0].binary
+  #   IO.copy_stream res.images[0], "rocket.png"
   class Images
     include Format
 
@@ -30,7 +30,7 @@ class LLM::Gemini
     # @example
     #   llm = LLM.gemini(ENV["KEY"])
     #   res = llm.images.create prompt: "A dog on a rocket to the moon"
-    #   File.binwrite "rocket.png", res.images[0].binary
+    #   IO.copy_stream res.images[0], "rocket.png"
     # @see https://ai.google.dev/gemini-api/docs/image-generation Gemini docs
     # @param [String] prompt The prompt
     # @param [Hash] params Other parameters (see Gemini docs)
@@ -56,7 +56,7 @@ class LLM::Gemini
     # @example
     #   llm = LLM.gemini(ENV["KEY"])
     #   res = llm.images.edit image: LLM::File("cat.png"), prompt: "Add a hat to the cat"
-    #   File.binwrite "hatoncat.png", res.images[0].binary
+    #   IO.copy_stream res.images[0], "hatoncat.png"
     # @see https://ai.google.dev/gemini-api/docs/image-generation Gemini docs
     # @param [LLM::File] image The image to edit
     # @param [String] prompt The prompt

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

@@ -0,0 +1,69 @@
+# frozen_string_literal: true
+
+class LLM::Gemini
+  ##
+  # The {LLM::Gemini::Models LLM::Gemini::Models} class provides a model
+  # object for interacting with [Gemini's models API](https://ai.google.dev/api/models?hl=en#method:-models.list).
+  # The models API allows a client to query Gemini for a list of models
+  # that are available for use with the Gemini API.
+  #
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm = LLM.gemini(ENV["KEY"])
+  #   res = llm.models.all
+  #   res.each do |model|
+  #     print "id: ", model.id, "\n"
+  #   end
+  class Models
+    include LLM::Utils
+
+    ##
+    # Returns a new Models object
+    # @param provider [LLM::Provider]
+    # @return [LLM::Gemini::Models]
+    def initialize(provider)
+      @provider = provider
+    end
+
+    ##
+    # List all models
+    # @example
+    #   llm = LLM.gemini(ENV["KEY"])
+    #   res = llm.models.all
+    #   res.each do |model|
+    #     print "id: ", model.id, "\n"
+    #   end
+    # @see https://ai.google.dev/api/models?hl=en#method:-models.list Gemini docs
+    # @param [Hash] params Other parameters (see Gemini docs)
+    # @raise (see LLM::Provider#request)
+    # @return [LLM::Response::ModelList]
+    def all(**params)
+      query = URI.encode_www_form(params.merge!(key: secret))
+      req = Net::HTTP::Get.new("/v1beta/models?#{query}", headers)
+      res = request(http, req)
+      LLM::Response::ModelList.new(res).tap { |modellist|
+        models = modellist.body["models"].map do |model|
+          model = model.transform_keys { snakecase(_1) }
+          LLM::Model.from_hash(model).tap { _1.provider = @provider }
+        end
+        modellist.models = models
+      }
+    end
+
+    private
+
+    def http
+      @provider.instance_variable_get(:@http)
+    end
+
+    def secret
+      @provider.instance_variable_get(:@secret)
+    end
+
+    [:headers, :request].each do |m|
+      define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
+    end
+  end
+end

data/lib/llm/providers/gemini/response_parser.rb

@@ -44,11 +44,7 @@ class LLM::Gemini
         images: body["candidates"].flat_map do |candidate|
           candidate["content"]["parts"].filter_map do
             next unless _1.dig("inlineData", "data")
-            OpenStruct.from_hash(
-              mime_type: _1["inlineData"]["mimeType"],
-              encoded: _1["inlineData"]["data"],
-              binary: _1["inlineData"]["data"].unpack1("m0")
-            )
+            StringIO.new(_1["inlineData"]["data"].unpack1("m0"))
           end
         end
       }

data/lib/llm/providers/gemini.rb

@@ -34,6 +34,7 @@ module LLM
     require_relative "gemini/images"
     require_relative "gemini/files"
     require_relative "gemini/audio"
+    require_relative "gemini/models"
     include Format
 
     HOST = "generativelanguage.googleapis.com"
@@ -52,6 +53,7 @@ module LLM
     # @raise (see LLM::Provider#request)
     # @return (see LLM::Provider#embed)
     def embed(input, model: "text-embedding-004", **params)
+      model = model.respond_to?(:id) ? model.id : model
       path = ["/v1beta/models/#{model}", "embedContent?key=#{@secret}"].join(":")
       req = Net::HTTP::Post.new(path, headers)
       req.body = JSON.dump({content: {parts: [{text: input}]}})
@@ -65,15 +67,19 @@ module LLM
     # @param prompt (see LLM::Provider#complete)
     # @param role (see LLM::Provider#complete)
     # @param model (see LLM::Provider#complete)
+    # @param schema (see LLM::Provider#complete)
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
+    # @raise [LLM::Error::PromptError]
+    #  When given an object a provider does not understand
     # @return (see LLM::Provider#complete)
-    def complete(prompt, role = :user, model: "gemini-1.5-flash", **params)
+    def complete(prompt, role = :user, model: default_model, schema: nil, **params)
+      model.respond_to?(:id) ? model.id : model
       path = ["/v1beta/models/#{model}", "generateContent?key=#{@secret}"].join(":")
       req  = Net::HTTP::Post.new(path, headers)
       messages = [*(params.delete(:messages) || []), LLM::Message.new(role, prompt)]
-      body = JSON.dump({contents: format(messages)}).b
+      body = JSON.dump({contents: format(messages)}.merge!(expand_schema(schema)))
       set_body_stream(req, StringIO.new(body))
       res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
@@ -101,6 +107,13 @@ module LLM
       LLM::Gemini::Files.new(self)
     end
 
+    ##
+    # Provides an interface to Gemini's models API
+    # @see https://ai.google.dev/gemini-api/docs/models Gemini docs
+    def models
+      LLM::Gemini::Models.new(self)
+    end
+
     ##
     # @return (see LLM::Provider#assistant_role)
     def assistant_role
@@ -108,9 +121,11 @@ module LLM
     end
 
     ##
-    # @return (see LLM::Provider#models)
-    def models
-      @models ||= load_models!("gemini")
+    # Returns the default model for chat completions
+    # @see https://ai.google.dev/gemini-api/docs/models#gemini-1.5-flash gemini-1.5-flash
+    # @return [String]
+    def default_model
+      "gemini-1.5-flash"
     end
 
     private
@@ -121,6 +136,16 @@ module LLM
       }
     end
 
+    def expand_schema(schema)
+      return {} unless schema
+      {
+        "generationConfig" => {
+          "response_mime_type" => "application/json",
+          "response_schema" => schema
+        }
+      }
+    end
+
     def response_parser
       LLM::Gemini::ResponseParser
     end

data/lib/llm/providers/ollama/format.rb

@@ -28,14 +28,22 @@ class LLM::Ollama
     # @return [String, Hash]
     #  The formatted content
     def format_content(content)
-      if LLM::File === content
+      case content
+      when LLM::File
         if content.image?
           {content: "This message has an image associated with it", images: [content.to_b64]}
         else
-          raise TypeError, "'#{content.path}' was not recognized as an image file."
+          raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                         "is not an image, and therefore not supported by the " \
+                                         "Ollama API"
         end
-      else
+      when String
         {content:}
+      when LLM::Message
+        format_content(content.content)
+      else
+        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                       "is not supported by the Ollama API"
       end
     end
   end

data/lib/llm/providers/ollama/models.rb

@@ -0,0 +1,66 @@
+# frozen_string_literal: true
+
+class LLM::Ollama
+  ##
+  # The {LLM::Ollama::Models LLM::Ollama::Models} class provides a model
+  # object for interacting with [Ollama's models API](https://github.com/ollama/ollama/blob/main/docs/api.md#list-local-models).
+  # The models API allows a client to query Ollama for a list of models
+  # that are available for use with the Ollama API.
+  #
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm = LLM.ollama(nil)
+  #   res = llm.models.all
+  #   res.each do |model|
+  #     print "id: ", model.id, "\n"
+  #   end
+  class Models
+    include LLM::Utils
+
+    ##
+    # Returns a new Models object
+    # @param provider [LLM::Provider]
+    # @return [LLM::Ollama::Models]
+    def initialize(provider)
+      @provider = provider
+    end
+
+    ##
+    # List all models
+    # @example
+    #   llm = LLM.ollama(nil)
+    #   res = llm.models.all
+    #   res.each do |model|
+    #     print "id: ", model.id, "\n"
+    #   end
+    # @see https://github.com/ollama/ollama/blob/main/docs/api.md#list-local-models Ollama docs
+    # @see https://ollama.com/library Ollama library
+    # @param [Hash] params Other parameters (see Ollama docs)
+    # @raise (see LLM::Provider#request)
+    # @return [LLM::Response::ModelList]
+    def all(**params)
+      query = URI.encode_www_form(params)
+      req = Net::HTTP::Get.new("/api/tags?#{query}", headers)
+      res = request(http, req)
+      LLM::Response::ModelList.new(res).tap { |modellist|
+        models = modellist.body["models"].map do |model|
+          model = model.transform_keys { snakecase(_1) }
+          LLM::Model.from_hash(model).tap { _1.provider = @provider }
+        end
+        modellist.models = models
+      }
+    end
+
+    private
+
+    def http
+      @provider.instance_variable_get(:@http)
+    end
+
+    [:headers, :request].each do |m|
+      define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
+    end
+  end
+end

data/lib/llm/providers/ollama.rb

@@ -22,6 +22,7 @@ module LLM
     require_relative "ollama/error_handler"
     require_relative "ollama/response_parser"
     require_relative "ollama/format"
+    require_relative "ollama/models"
     include Format
 
     HOST = "localhost"
@@ -56,16 +57,30 @@ module LLM
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
+    # @raise [LLM::Error::PromptError]
+    #  When given an object a provider does not understand
     # @return (see LLM::Provider#complete)
-    def complete(prompt, role = :user, model: "llama3.2", **params)
-      params   = {model:, stream: false}.merge!(params)
-      req      = Net::HTTP::Post.new("/api/chat", headers)
+    def complete(prompt, role = :user, model: default_model, schema: nil, **params)
+      params = {model:, stream: false}
+                 .merge!(expand_schema(schema))
+                 .merge!(params)
+                 .compact
+      req = Net::HTTP::Post.new("/api/chat", headers)
       messages = [*(params.delete(:messages) || []), LLM::Message.new(role, prompt)]
-      req.body = JSON.dump({messages: format(messages)}.merge!(params))
-      res      = request(@http, req)
+      body = JSON.dump({messages: format(messages)}.merge!(params))
+      set_body_stream(req, StringIO.new(body))
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # Provides an interface to Ollama's models API
+    # @see https://github.com/ollama/ollama/blob/main/docs/api.md#list-local-models Ollama docs
+    # @return [LLM::Ollama::Models]
+    def models
+      LLM::Ollama::Models.new(self)
+    end
+
     ##
     # @return (see LLM::Provider#assistant_role)
     def assistant_role
@@ -73,9 +88,11 @@ module LLM
     end
 
     ##
-    # @return (see LLM::Provider#models)
-    def models
-      @models ||= load_models!("ollama")
+    # Returns the default model for chat completions
+    # @see https://ollama.com/library llama3.2
+    # @return [String]
+    def default_model
+      "llama3.2"
     end
 
     private
@@ -87,6 +104,11 @@ module LLM
       }
     end
 
+    def expand_schema(schema)
+      return {} unless schema
+      {format: schema}
+    end
+
     def response_parser
       LLM::Ollama::ResponseParser
     end

data/lib/llm/providers/openai/audio.rb

@@ -9,8 +9,6 @@ class LLM::OpenAI
   #   res = llm.audio.create_speech(input: "A dog on a rocket to the moon")
   #   File.binwrite("rocket.mp3", res.audio.string)
   class Audio
-    require "stringio"
-
     ##
     # Returns a new Audio object
     # @param provider [LLM::Provider]

data/lib/llm/providers/openai/format.rb

@@ -42,7 +42,12 @@ class LLM::OpenAI
       when URI
         [{type: :image_url, image_url: {url: content.to_s}}]
       when LLM::File
-        [{type: :image_url, image_url: {url: content.to_data_uri}}]
+        file = content
+        if file.image?
+          [{type: :image_url, image_url: {url: file.to_data_uri}}]
+        else
+          [{type: :file, file: {filename: file.basename, file_data: file.to_data_uri}}]
+        end
       when LLM::Response::File
         [{type: :file, file: {file_id: content.id}}]
       when String

data/lib/llm/providers/openai/images.rb

@@ -24,7 +24,7 @@ class LLM::OpenAI
   #   llm = LLM.openai(ENV["KEY"])
   #   res = llm.images.create prompt: "A dog on a rocket to the moon",
   #                           response_format: "b64_json"
-  #   File.binwrite("rocket.png", res.images[0].binary)
+  #   IO.copy_stream res.images[0], "rocket.png"
   class Images
     ##
     # Returns a new Images object

data/lib/llm/providers/openai/models.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+class LLM::OpenAI
+  ##
+  # The {LLM::OpenAI::Models LLM::OpenAI::Models} class provides a model
+  # object for interacting with [OpenAI's models API](https://platform.openai.com/docs/api-reference/models/list).
+  # The models API allows a client to query OpenAI for a list of models
+  # that are available for use with the OpenAI API.
+  #
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm = LLM.openai(ENV["KEY"])
+  #   res = llm.models.all
+  #   res.each do |model|
+  #     print "id: ", model.id, "\n"
+  #   end
+  class Models
+    ##
+    # Returns a new Models object
+    # @param provider [LLM::Provider]
+    # @return [LLM::OpenAI::Files]
+    def initialize(provider)
+      @provider = provider
+    end
+
+    ##
+    # List all models
+    # @example
+    #   llm = LLM.openai(ENV["KEY"])
+    #   res = llm.models.all
+    #   res.each do |model|
+    #     print "id: ", model.id, "\n"
+    #   end
+    # @see https://platform.openai.com/docs/api-reference/models/list OpenAI docs
+    # @param [Hash] params Other parameters (see OpenAI docs)
+    # @raise (see LLM::Provider#request)
+    # @return [LLM::Response::FileList]
+    def all(**params)
+      query = URI.encode_www_form(params)
+      req = Net::HTTP::Get.new("/v1/models?#{query}", headers)
+      res = request(http, req)
+      LLM::Response::ModelList.new(res).tap { |modellist|
+        models = modellist.body["data"].map do |model|
+          LLM::Model.from_hash(model).tap { _1.provider = @provider }
+        end
+        modellist.models = models
+      }
+    end
+
+    private
+
+    def http
+      @provider.instance_variable_get(:@http)
+    end
+
+    [:headers, :request, :set_body_stream].each do |m|
+      define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
+    end
+  end
+end

data/lib/llm/providers/openai/response_parser.rb

@@ -69,11 +69,7 @@ class LLM::OpenAI
         urls: body["data"].filter_map { _1["url"] },
         images: body["data"].filter_map do
           next unless _1["b64_json"]
-          OpenStruct.from_hash(
-            mime_type: nil,
-            encoded: _1["b64_json"],
-            binary: _1["b64_json"].unpack1("m0")
-          )
+          StringIO.new(_1["b64_json"].unpack1("m0"))
         end
       }
     end

data/lib/llm/providers/openai/responses.rb

@@ -49,13 +49,19 @@ class LLM::OpenAI
     # @param model (see LLM::Provider#complete)
     # @param [Hash] params Response params
     # @raise (see LLM::Provider#request)
+    # @raise [LLM::Error::PromptError]
+    #  When given an object a provider does not understand
     # @return [LLM::Response::Output]
-    def create(prompt, role = :user, model: "gpt-4o-mini", **params)
-      params   = {model:}.merge!(params)
-      req      = Net::HTTP::Post.new("/v1/responses", headers)
+    def create(prompt, role = :user, model: @provider.default_model, schema: nil, **params)
+      params = {model:}
+                 .merge!(expand_schema(schema))
+                 .merge!(params)
+                 .compact
+      req = Net::HTTP::Post.new("/v1/responses", headers)
       messages = [*(params.delete(:input) || []), LLM::Message.new(role, prompt)]
-      req.body = JSON.dump({input: format(messages, :response)}.merge!(params))
-      res      = request(http, req)
+      body = JSON.dump({input: format(messages, :response)}.merge!(params))
+      set_body_stream(req, StringIO.new(body))
+      res = request(http, req)
       LLM::Response::Output.new(res).extend(response_parser)
     end
 
@@ -92,7 +98,7 @@ class LLM::OpenAI
       @provider.instance_variable_get(:@http)
     end
 
-    [:response_parser, :headers, :request].each do |m|
+    [:response_parser, :headers, :request, :set_body_stream, :expand_schema].each do |m|
       define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
     end
   end