llm.rb 0.3.2 → 0.3.3

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]

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 +49,17 @@ 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
 
@@ -92,7 +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

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "0.3.2"
+  VERSION = "0.3.3"
 end

data/lib/llm.rb

@@ -1,6 +1,8 @@
 # frozen_string_literal: true
 
 module LLM
+  require "stringio"
+  require_relative "llm/core_ext/ostruct"
   require_relative "llm/version"
   require_relative "llm/utils"
   require_relative "llm/error"
@@ -13,7 +15,6 @@ module LLM
   require_relative "llm/provider"
   require_relative "llm/chat"
   require_relative "llm/buffer"
-  require_relative "llm/core_ext/ostruct"
 
   module_function
 

data/spec/anthropic/completion_spec.rb

@@ -42,6 +42,42 @@ RSpec.describe "LLM::Anthropic: completions" do
     end
   end
 
+  context "when given a URI to an image",
+          vcr: {cassette_name: "anthropic/completions/successful_response_uri_image"} do
+    subject { response.choices[0].content.downcase[0..2] }
+    let(:response) do
+      anthropic.complete([
+        "Is this image the flag of brazil ? ",
+        "Answer with yes or no. ",
+        "Nothing else.",
+        uri
+      ], :user)
+    end
+    let(:uri) { URI("https://upload.wikimedia.org/wikipedia/en/thumb/0/05/Flag_of_Brazil.svg/250px-Flag_of_Brazil.svg.png") }
+
+    it "describes the image" do
+      is_expected.to eq("yes")
+    end
+  end
+
+  context "when given a local reference to an image",
+          vcr: {cassette_name: "anthropic/completions/successful_response_file_image"} do
+    subject { response.choices[0].content.downcase[0..2] }
+    let(:response) do
+      anthropic.complete([
+        "Is this image a representation of a blue book ?",
+        "Answer with yes or no.",
+        "Nothing else.",
+        file
+      ], :user)
+    end
+    let(:file) { LLM::File("spec/fixtures/images/bluebook.png") }
+
+    it "describes the image" do
+      is_expected.to eq("yes")
+    end
+  end
+
   context "when given an unauthorized response",
           vcr: {cassette_name: "anthropic/completions/unauthorized_response"} do
     subject(:response) { anthropic.complete("Hello", :user) }

data/spec/anthropic/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Anthropic::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.anthropic(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "anthropic/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/gemini/images_spec.rb

@@ -14,12 +14,8 @@ RSpec.describe "LLM::Gemini::Images" do
       expect(response).to be_instance_of(LLM::Response::Image)
     end
 
-    it "returns an encoded string" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns a binary string" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 
@@ -36,12 +32,8 @@ RSpec.describe "LLM::Gemini::Images" do
       expect(response).to be_instance_of(LLM::Response::Image)
     end
 
-    it "returns data" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns a url" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 end

data/spec/gemini/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Gemini::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.gemini(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "gemini/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/llm/conversation_spec.rb

@@ -27,7 +27,7 @@ RSpec.describe "LLM::Chat: non-lazy" do
         bot = nil
         inputs.zip(outputs).each_with_index do |(input, output), index|
           expect(provider).to receive(:complete)
-                                .with(input.content, instance_of(Symbol), messages:)
+                                .with(input.content, instance_of(Symbol), messages:, model: provider.default_model)
                                 .and_return(OpenStruct.new(choices: [output]))
           bot = index.zero? ? provider.chat!(input.content, :system) : bot.chat(input.content)
           messages.concat([input, output])
@@ -117,7 +117,8 @@ RSpec.describe "LLM::Chat: lazy" do
 
       context "when given a specific model",
               vcr: {cassette_name: "openai/lazy_conversation/completions/successful_response_o3_mini"} do
-        let(:conversation) { described_class.new(provider, model: provider.models["o3-mini"]).lazy }
+        let(:model) { provider.models.all.find { _1.id == "o3-mini" } }
+        let(:conversation) { described_class.new(provider, model:).lazy }
 
         it "maintains the model throughout a conversation" do
           conversation.chat(prompt, :system)
@@ -179,7 +180,8 @@ RSpec.describe "LLM::Chat: lazy" do
 
       context "when given a specific model",
               vcr: {cassette_name: "openai/lazy_conversation/responses/successful_response_o3_mini"} do
-        let(:conversation) { described_class.new(provider, model: provider.models["o3-mini"]).lazy }
+        let(:model) { provider.models.all.find { _1.id == "o3-mini" } }
+        let(:conversation) { described_class.new(provider, model:).lazy }
 
         it "maintains the model throughout a conversation" do
           conversation.respond(prompt, :developer)

data/spec/ollama/models_spec.rb

@@ -0,0 +1,20 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Ollama::Models" do
+  let(:provider) { LLM.ollama(nil, host: "eel.home.network") }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "ollama/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/openai/completion_spec.rb

@@ -64,6 +64,25 @@ RSpec.describe "LLM::OpenAI: completions" do
     end
   end
 
+  context "when asked to describe an audio file",
+          vcr: {cassette_name: "openai/completions/describe_pdf_document"} do
+    let(:file) { LLM::File("spec/fixtures/documents/freebsd.sysctl.pdf") }
+    let(:response) do
+      openai.complete([
+        "This PDF document describes sysctl nodes on FreeBSD",
+        "Answer yes or no.",
+        "Nothing else",
+        file
+      ], :user)
+    end
+
+    subject { response.choices[0].content.downcase[0..2] }
+
+    it "is successful" do
+      is_expected.to eq("yes")
+    end
+  end
+
   context "when given a 'bad request' response",
           vcr: {cassette_name: "openai/completions/bad_request"} do
     subject(:response) { openai.complete(URI("/foobar.exe"), :user) }

data/spec/openai/images_spec.rb

@@ -40,12 +40,8 @@ RSpec.describe "LLM::OpenAI::Images" do
       expect(response.images).to be_instance_of(Array)
     end
 
-    it "returns an encoded string" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns an binary string" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 

data/spec/openai/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::OpenAI::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.openai(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "openai/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end