llm.rb 0.3.1 → 0.3.3

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}]}})
@@ -68,13 +70,18 @@ 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: "gemini-1.5-flash", **params)
-      path     = ["/v1beta/models/#{model}", "generateContent?key=#{@secret}"].join(":")
-      req      = Net::HTTP::Post.new(path, headers)
+    def complete(prompt, role = :user, model: default_model, **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)]
-      req.body = JSON.dump({contents: format(messages)})
-      res      = request(@http, req)
+      body = JSON.dump({contents: format(messages)})
+      set_body_stream(req, StringIO.new(body))
+
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
@@ -100,6 +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
@@ -107,9 +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

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 +57,28 @@ 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, **params)
+      params = {model:, stream: false}.merge!(params)
+      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 +86,11 @@ 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

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]
@@ -57,7 +55,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file:, model:))
       req = Net::HTTP::Post.new("/v1/audio/transcriptions", headers)
       req["content-type"] = multi.content_type
-      req.body_stream = multi.body
+      set_body_stream(req, multi.body)
       res = request(http, req)
       LLM::Response::AudioTranscription.new(res).tap { _1.text = _1.body["text"] }
     end
@@ -79,7 +77,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file:, model:))
       req = Net::HTTP::Post.new("/v1/audio/translations", headers)
       req["content-type"] = multi.content_type
-      req.body_stream = multi.body
+      set_body_stream(req, multi.body)
       res = request(http, req)
       LLM::Response::AudioTranslation.new(res).tap { _1.text = _1.body["text"] }
     end
@@ -90,7 +88,7 @@ class LLM::OpenAI
       @provider.instance_variable_get(:@http)
     end
 
-    [:headers, :request].each do |m|
+    [:headers, :request, :set_body_stream].each do |m|
       define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
     end
   end

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

@@ -73,7 +73,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file:, purpose:))
       req = Net::HTTP::Post.new("/v1/files", headers)
       req["content-type"] = multi.content_type
-      req.body_stream = multi.body
+      set_body_stream(req, multi.body)
       res = request(http, req)
       LLM::Response::File.new(res)
     end
@@ -141,7 +141,7 @@ class LLM::OpenAI
       @provider.instance_variable_get(:@http)
     end
 
-    [:headers, :request].each do |m|
+    [:headers, :request, :set_body_stream].each do |m|
       define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
     end
   end

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

@@ -29,19 +29,55 @@ class LLM::OpenAI
     #  The formatted content
     def format_content(content, mode)
       if mode == :complete
-        case content
-        when Array then content.flat_map { format_content(_1, mode) }
-        when URI then [{type: :image_url, image_url: {url: content.to_s}}]
-        when LLM::Response::File then [{type: :file, file: {file_id: content.id}}]
-        else [{type: :text, text: content.to_s}]
-        end
+        format_complete(content)
       elsif mode == :response
-        case content
-        when Array then content.flat_map { format_content(_1, mode) }
-        when URI then [{type: :image_url, image_url: {url: content.to_s}}]
-        when LLM::Response::File then [{type: :input_file, file_id: content.id}]
-        else [{type: :input_text, text: content.to_s}]
+        format_response(content)
+      end
+    end
+
+    def format_complete(content)
+      case content
+      when Array
+        content.flat_map { format_complete(_1) }
+      when URI
+        [{type: :image_url, image_url: {url: content.to_s}}]
+      when LLM::File
+        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
+        [{type: :text, text: content.to_s}]
+      when LLM::Message
+        format_complete(content.content)
+      else
+        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                       "is not supported by the OpenAI chat completions API"
+      end
+    end
+
+    def format_response(content)
+      case content
+      when Array
+        content.flat_map { format_response(_1) }
+      when LLM::Response::File
+        file = LLM::File(content.filename)
+        if file.image?
+          [{type: :input_image, file_id: content.id}]
+        else
+          [{type: :input_file, file_id: content.id}]
         end
+      when String
+        [{type: :input_text, text: content.to_s}]
+      when LLM::Message
+        format_response(content.content)
+      else
+        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                       "is not supported by the OpenAI responses API"
       end
     end
   end

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
@@ -69,7 +69,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(image:, model:))
       req = Net::HTTP::Post.new("/v1/images/variations", headers)
       req["content-type"] = multi.content_type
-      req.body_stream = multi.body
+      set_body_stream(req, multi.body)
       res = request(http, req)
       LLM::Response::Image.new(res).extend(response_parser)
     end
@@ -91,7 +91,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(image:, prompt:, model:))
       req = Net::HTTP::Post.new("/v1/images/edits", headers)
       req["content-type"] = multi.content_type
-      req.body_stream = multi.body
+      set_body_stream(req, multi.body)
       res = request(http, req)
       LLM::Response::Image.new(res).extend(response_parser)
     end
@@ -102,7 +102,7 @@ class LLM::OpenAI
       @provider.instance_variable_get(:@http)
     end
 
-    [:response_parser, :headers, :request].each do |m|
+    [:response_parser, :headers, :request, :set_body_stream].each do |m|
       define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
     end
   end

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

@@ -16,6 +16,20 @@ class LLM::OpenAI
   #   res1 = llm.responses.create "Your task is to help me with math", :developer
   #   res2 = llm.responses.create "5 + 5  = ?", :user, previous_response_id: res1.id
   #   [res1,res2].each { llm.responses.delete(_1) }
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm  = LLM.openai(ENV["KEY"])
+  #   file = llm.files.create file: LLM::File("/images/hat.png")
+  #   res  = llm.responses.create ["Describe the image", file]
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm  = LLM.openai(ENV["KEY"])
+  #   file = llm.files.create file: LLM::File("/documents/freebsd.pdf")
+  #   res  = llm.responses.create ["Describe the document, file]
   class Responses
     include Format
 
@@ -35,13 +49,17 @@ 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, **params)
+      params = {model:}.merge!(params)
+      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
 
@@ -78,7 +96,7 @@ class LLM::OpenAI
       @provider.instance_variable_get(:@http)
     end
 
-    [:response_parser, :headers, :request].each do |m|
+    [:response_parser, :headers, :request, :set_body_stream].each do |m|
       define_method(m) { |*args, &b| @provider.send(m, *args, &b) }
     end
   end

data/lib/llm/providers/openai.rb

@@ -12,6 +12,7 @@ module LLM
     require_relative "openai/images"
     require_relative "openai/audio"
     require_relative "openai/files"
+    require_relative "openai/models"
     include Format
 
     HOST = "api.openai.com"
@@ -46,13 +47,17 @@ 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: "gpt-4o-mini", **params)
-      params   = {model:}.merge!(params)
-      req      = Net::HTTP::Post.new("/v1/chat/completions", headers)
+    def complete(prompt, role = :user, model: default_model, **params)
+      params = {model:}.merge!(params)
+      req = Net::HTTP::Post.new("/v1/chat/completions", headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      req.body = JSON.dump({messages: format(messages, :complete)}.merge!(params))
-      res      = request(@http, req)
+      body = JSON.dump({messages: format(messages, :complete)}.merge!(params))
+      set_body_stream(req, StringIO.new(body))
+
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
@@ -88,14 +93,26 @@ module LLM
       LLM::OpenAI::Files.new(self)
     end
 
+    ##
+    # Provides an interface to OpenAI's models API
+    # @see https://platform.openai.com/docs/api-reference/models/list OpenAI docs
+    # @return [LLM::OpenAI::Models]
+    def models
+      LLM::OpenAI::Models.new(self)
+    end
+
     ##
     # @return (see LLM::Provider#assistant_role)
     def assistant_role
       "assistant"
     end
 
-    def models
-      @models ||= load_models!("openai")
+    ##
+    # Returns the default model for chat completions
+    # @see https://platform.openai.com/docs/models/gpt-4o-mini gpt-4o-mini
+    # @return [String]
+    def default_model
+      "gpt-4o-mini"
     end
 
     private

data/lib/llm/response/modellist.rb

@@ -0,0 +1,18 @@
+# frozen_string_literal: true
+
+module LLM
+  ##
+  # The {LLM::Response::ModelList LLM::Response::ModelList} class represents a
+  # list of model objects that are returned by a provider. It is an Enumerable
+  # object, and can be used to iterate over the model objects in a way that is
+  # similar to an array. Each element is an instance of OpenStruct.
+  class Response::ModelList < Response
+    include Enumerable
+
+    attr_accessor :models
+
+    def each(&)
+      @models.each(&)
+    end
+  end
+end

data/lib/llm/response.rb

@@ -13,6 +13,7 @@ module LLM
     require_relative "response/file"
     require_relative "response/filelist"
     require_relative "response/download_file"
+    require_relative "response/modellist"
 
     ##
     # @param [Net::HTTPResponse] res