llm.rb 0.3.1 → 0.3.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 3939075c064b4abfd8853c3f67b6db7df6111d340d658d4d8ad0c4d1bccc96bc
-  data.tar.gz: 0ca274d3e4b032c25730aef896df903681c28033ebb0907c965339a33aff56d1
+  metadata.gz: c8ce8caa7c769da9197528a864c153071f3c4aca15718efc985e543911c04ce2
+  data.tar.gz: 389ff41e9e2b35782b1484048b7597f5573bf2a86cf9eaff8cfd7c4cb2b19be3
 SHA512:
-  metadata.gz: feaf87457b8fa5b4f756a5fe8cc1f670c8b0286a730fe00273bc99678092fe7f704d58f01ba0a0baf4072a0dcee063bc87cf88bc7cdf53125334476adbce41f6
-  data.tar.gz: 3be8b460d9b483c0e172d9159b2394ea39da7a1475aee3ab47b224303e2a251f3b04f0543402494485040998225f84342be986db8c7b8ea80df92f561d4d6d92
+  metadata.gz: 39c8f71eae878b5787ca839138de07ce06cba4fdee0bafb1bd75a71f3b3e59ee08fa05f5d9f280522ec751722c9a8a15430d1b999b05e14052d85c745bf9781c
+  data.tar.gz: fdb5268b0095f09b41481e6c7071a2dae66cf9a3fd21755834b76040e13a236bc18122dfe95230dd49a827e774487a1add1aa35c8976f284a03afad881321b46

data/README.md

@@ -26,6 +26,7 @@ llm = LLM.openai("yourapikey")
 llm = LLM.gemini("yourapikey")
 llm = LLM.anthropic("yourapikey")
 llm = LLM.ollama(nil)
+llm = LLM.voyageai("yourapikey")
 ```
 
 ### Conversations
@@ -37,7 +38,9 @@ The following example enables lazy mode for a
 object by entering into a "lazy" conversation where messages are buffered and
 sent to the provider only when necessary.  Both lazy and non-lazy conversations
 maintain a message thread that can be reused as context throughout a conversation.
-The example uses the stateless chat completions API that all LLM providers support:
+The example captures the spirit of llm.rb by demonstrating how objects cooperate
+together through composition, and it uses the stateless chat completions API that
+all LLM providers support:
 
 ```ruby
 #!/usr/bin/env ruby
@@ -122,13 +125,10 @@ for more information on how to use the audio generation API:
 ```ruby
 #!/usr/bin/env ruby
 require "llm"
-require "open-uri"
-require "fileutils"
 
 llm = LLM.openai(ENV["KEY"])
 res = llm.audio.create_speech(input: "Hello world")
-File.binwrite File.join(Dir.home, "hello.mp3"),
-	          res.audio.string
+IO.copy_stream res.audio, File.join(Dir.home, "hello.mp3")
 ```
 
 #### Transcribe
@@ -151,8 +151,6 @@ examples and documentation
 ```ruby
 #!/usr/bin/env ruby
 require "llm"
-require "open-uri"
-require "fileutils"
 
 llm = LLM.openai(ENV["KEY"])
 res = llm.audio.create_transcription(
@@ -180,9 +178,8 @@ examples and documentation
 
 
 ```ruby
+#!/usr/bin/env ruby
 require "llm"
-require "open-uri"
-require "fileutils"
 
 llm = LLM.openai(ENV["KEY"])
 res = llm.audio.create_translation(
@@ -320,6 +317,48 @@ bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n"
 #             contains information about the features, installation, and usage of OpenBSD.
 ```
 
+### Prompts
+
+#### Multimodal
+
+Generally all providers accept text prompts but some providers can
+also understand URLs, and various file types (eg images, audio, video,
+etc). The llm.rb approach to multimodal prompts is to let you pass `URI`
+objects to describe links, `LLM::File` / `LLM::Response::File` objects
+to describe files, `String` objects to describe text blobs, or an array
+of the forementioned objects to describe multiple objects in a single
+prompt. Each object is a first class citizen that can be passed directly
+to a prompt.
+
+For more depth and examples on how to use the multimodal API, please see
+the [provider-specific documentation](https://0x1eef.github.io/x/llm.rb/)
+for more provider-specific examples – there can be subtle differences
+between providers and even between APIs from the same provider that are
+not covered in the README:
+
+```ruby
+#!/usr/bin/env ruby
+require "llm"
+
+llm = LLM.openai(ENV["KEY"])
+bot = LLM::Chat.new(llm).lazy
+
+bot.chat URI("https://example.com/path/to/image.png")
+bot.chat "Describe the above image"
+bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n" }
+
+file = bot.files.create(file: LLM::File("/documents/openbsd_is_awesome.pdf"))
+bot.chat file
+bot.chat "What is this file about?"
+bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n" }
+
+bot.chat [LLM::File("/images/puffy.png"), "What is this image about?"]
+bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n" }
+
+bot.chat [LLM::File("/images/beastie.png"), "What is this image about?"]
+bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n" }
+```
+
 ### Embeddings
 
 #### Text
@@ -350,6 +389,38 @@ print res.embeddings[0].size, "\n"
 # 1536
 ```
 
+### Models
+
+#### List
+
+Almost all LLM providers provide a models endpoint that allows a client to
+query the list of models that are available to use. The list is dynamic,
+maintained by LLM providers, and it is independent of a specific llm.rb release.
+True to the llm.rb spirit of small, composable objects that cooperate with
+each other, a
+[LLM::Model](https://0x1eef.github.io/x/llm.rb/LLM/Model.html)
+object can be used instead of a string that describes a model name (although
+either works). Let's take a look at an example:
+
+```ruby
+#!/usr/bin/env ruby
+require "llm"
+
+##
+# List all models
+llm = LLM.openai(ENV["KEY"])
+llm.models.all.each do |model|
+  print "model: ", model.id, "\n"
+end
+
+##
+# Select a model
+model = llm.models.all.find { |m| m.id == "gpt-3.5-turbo" }
+bot = LLM::Chat.new(llm, model:)
+bot.chat "Hello #{model.id} :)"
+bot.messages.select(&:assistant?).each { print "[#{_1.role}] ", _1.content, "\n" }
+```
+
 ### Memory
 
 #### Child process
@@ -372,7 +443,7 @@ llm = LLM.gemini(ENV["KEY"])
 fork do
   %w[dog cat sheep goat capybara].each do |animal|
     res = llm.images.create(prompt: "a #{animal} on a rocket to the moon")
-    File.binwrite "#{animal}.png", res.images[0].binary
+    IO.copy_stream res.images[0], "#{animal}.png"
   end
 end
 Process.wait
@@ -394,19 +465,17 @@ llm.rb can be installed via rubygems.org:
 
 ## Philosophy
 
-llm.rb was built for developers who believe that simplicity is strength.
-It provides a clean, dependency-free interface to Large Language Models,
-treating Ruby itself as the primary platform – not Rails or any other
-specific framework or library. There is no hidden magic or extreme
-metaprogramming.
+llm.rb was built for developers who believe that simplicity can be challenging
+but it is always worth it. It provides a clean, dependency-free interface to
+Large Language Models, treating Ruby itself as the primary platform –
+not Rails or any other specific framework or library. There is no hidden
+magic or complex metaprogramming.
 
 Every part of llm.rb is designed to be explicit, composable, memory-safe,
 and production-ready without compromise. No unnecessary abstractions,
-no global configuration, and no dependencies that aren't part of standard
-Ruby. It has been inspired in part by other languages such as Python, but
-it is not a port of any other library.
-
-Good software doesn’t need marketing. It just needs to work. :)
+no global configuration, no global state, and no dependencies that aren't
+part of standard Ruby. It has been inspired in part by other languages such
+as Python, but it is not a port of any other library.
 
 ## License
 

data/lib/llm/chat.rb

@@ -27,11 +27,13 @@ module LLM
     ##
     # @param [LLM::Provider] provider
     #  A provider
+    # @param [String] model
+    #  The model to maintain throughout the conversation
     # @param [Hash] params
-    #  The parameters to maintain throughout the conversation
-    def initialize(provider, params = {})
+    #  Other parameters to maintain throughout the conversation
+    def initialize(provider, model: provider.default_model, **params)
       @provider = provider
-      @params = params
+      @params = params.merge!(model:)
       @lazy = false
       @messages = []
     end

data/lib/llm/core_ext/ostruct.rb

@@ -18,7 +18,7 @@ class OpenStruct
       hash_obj.each do |key, value|
         visited_object[key] = walk(value)
       end
-      OpenStruct.new(visited_object)
+      new(visited_object)
     end
 
     private

data/lib/llm/error.rb

@@ -4,8 +4,9 @@ module LLM
   ##
   # The superclass of all LLM errors
   class Error < RuntimeError
-    def initialize
+    def initialize(...)
       block_given? ? yield(self) : nil
+      super
     end
 
     ##
@@ -17,6 +18,10 @@ module LLM
       attr_accessor :response
     end
 
+    ##
+    # When a prompt is given an object that's not understood
+    PromptError = Class.new(Error)
+
     ##
     # HTTPUnauthorized
     Unauthorized = Class.new(ResponseError)

data/lib/llm/file.rb

@@ -7,13 +7,20 @@
 class LLM::File
   ##
   # @return [String]
-  #  Returns the path to a file
+  #  Returns the path to the file
   attr_reader :path
 
   def initialize(path)
     @path = path
   end
 
+  ##
+  # @return [String]
+  #  Returns basename of the file
+  def basename
+    File.basename(path)
+  end
+
   ##
   # @return [String]
   #  Returns the MIME type of the file
@@ -42,6 +49,13 @@ class LLM::File
     [File.binread(path)].pack("m0")
   end
 
+  ##
+  # @return [String]
+  #  Returns the file contents in base64 URL format
+  def to_data_uri
+    "data:#{mime_type};base64,#{to_b64}"
+  end
+
   ##
   # @return [File]
   #  Yields an IO object suitable to be streamed

data/lib/llm/model.rb

@@ -1,7 +1,32 @@
 # frozen_string_literal: true
 
-class LLM::Model < Struct.new(:name, :parameters, :description, :to_param, keyword_init: true)
+##
+# The {LLM::Model LLM::Model} class represents an LLM model that
+# is available to use. Its properties are delegated to the underlying
+# response body, and vary by provider.
+class LLM::Model < OpenStruct
+  ##
+  # Returns a subclass of {LLM::Provider LLM::Provider}
+  # @return [LLM::Provider]
+  attr_accessor :provider
+
+  ##
+  # Returns the model ID
+  # @return [String]
+  def id
+    case @provider.class.to_s
+    when "LLM::Ollama"
+      self["name"]
+    when "LLM::Gemini"
+      self["name"].sub(%r|\Amodels/|, "")
+    else
+      self["id"]
+    end
+  end
+
+  ##
+  # @return [String]
   def to_json(*)
-    to_param.to_json(*)
+    id.to_json(*)
   end
 end

data/lib/llm/provider.rb

@@ -4,16 +4,7 @@
 # The Provider class represents an abstract class for
 # LLM (Language Model) providers.
 #
-# @note
-#  This class is not meant to be instantiated directly.
-#  Instead, use one of the subclasses that implement
-#  the methods defined here.
-#
 # @abstract
-# @see LLM::Provider::OpenAI
-# @see LLM::Provider::Anthropic
-# @see LLM::Provider::Gemini
-# @see LLM::Provider::Ollama
 class LLM::Provider
   require "net/http"
 
@@ -53,7 +44,7 @@ class LLM::Provider
   # @raise [NotImplementedError]
   #  When the method is not implemented by a subclass
   # @return [LLM::Response::Embedding]
-  def embed(input, model:, **params)
+  def embed(input, model: nil, **params)
     raise NotImplementedError
   end
 
@@ -78,7 +69,7 @@ class LLM::Provider
   # @raise [NotImplementedError]
   #  When the method is not implemented by a subclass
   # @return [LLM::Response::Completion]
-  def complete(prompt, role = :user, model:, **params)
+  def complete(prompt, role = :user, model: default_model, **params)
     raise NotImplementedError
   end
 
@@ -94,8 +85,8 @@ class LLM::Provider
   #  Other completion parameters to maintain throughout a chat
   # @raise (see LLM::Provider#complete)
   # @return [LLM::Chat]
-  def chat(prompt, role = :user, model: nil, **params)
-    LLM::Chat.new(self, params).lazy.chat(prompt, role)
+  def chat(prompt, role = :user, model: default_model, **params)
+    LLM::Chat.new(self, **params.merge(model:)).lazy.chat(prompt, role)
   end
 
   ##
@@ -110,8 +101,8 @@ class LLM::Provider
   #  Other completion parameters to maintain throughout a chat
   # @raise (see LLM::Provider#complete)
   # @return [LLM::Chat]
-  def chat!(prompt, role = :user, model: nil, **params)
-    LLM::Chat.new(self, params).chat(prompt, role)
+  def chat!(prompt, role = :user, model: default_model, **params)
+    LLM::Chat.new(self, **params.merge(model:)).chat(prompt, role)
   end
 
   ##
@@ -126,8 +117,8 @@ class LLM::Provider
   #  Other completion parameters to maintain throughout a chat
   # @raise (see LLM::Provider#complete)
   # @return [LLM::Chat]
-  def respond(prompt, role = :user, model: nil, **params)
-    LLM::Chat.new(self, params).lazy.respond(prompt, role)
+  def respond(prompt, role = :user, model: default_model, **params)
+    LLM::Chat.new(self, **params.merge(model:)).lazy.respond(prompt, role)
   end
 
   ##
@@ -142,8 +133,8 @@ class LLM::Provider
   #  Other completion parameters to maintain throughout a chat
   # @raise (see LLM::Provider#complete)
   # @return [LLM::Chat]
-  def respond!(prompt, role = :user, model: nil, **params)
-    LLM::Chat.new(self, params).respond(prompt, role)
+  def respond!(prompt, role = :user, model: default_model, **params)
+    LLM::Chat.new(self, **params.merge(model:)).respond(prompt, role)
   end
 
   ##
@@ -178,6 +169,13 @@ class LLM::Provider
     raise NotImplementedError
   end
 
+  ##
+  # @return [LLM::OpenAI::Models]
+  #  Returns an interface to the models API
+  def models
+    raise NotImplementedError
+  end
+
   ##
   # @return [String]
   #  Returns the role of the assistant in the conversation.
@@ -187,9 +185,9 @@ class LLM::Provider
   end
 
   ##
-  # @return [Hash<String, LLM::Model>]
-  #  Returns a hash of available models
-  def models
+  # @return [String]
+  #  Returns the default model for chat completions
+  def default_model
     raise NotImplementedError
   end
 
@@ -248,15 +246,13 @@ class LLM::Provider
   end
 
   ##
-  # @param [String] provider
-  #  The name of the provider
-  # @return [Hash<String, Hash>]
-  def load_models!(provider)
-    require "yaml" unless defined?(YAML)
-    rootdir  = File.realpath File.join(__dir__, "..", "..")
-    sharedir = File.join(rootdir, "share", "llm")
-    provider = provider.gsub(/[^a-z0-9]/i, "")
-    yaml     = File.join(sharedir, "models", "#{provider}.yml")
-    YAML.safe_load_file(yaml).transform_values { LLM::Model.new(_1) }
+  # @param [Net::HTTPRequest] req
+  #  The request to set the body stream for
+  # @param [IO] io
+  #  The IO object to set as the body stream
+  # @return [void]
+  def set_body_stream(req, io)
+    req.body_stream = io
+    req["transfer-encoding"] = "chunked" unless req["content-length"]
   end
 end

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

@@ -26,13 +26,26 @@ class LLM::Anthropic
     # @return [String, Hash]
     #  The formatted content
     def format_content(content)
-      if URI === content
-        [{
-          type: :image,
-          source: {type: :base64, media_type: LLM::File(content.to_s).mime_type, data: [content.to_s].pack("m0")}
-        }]
+      case content
+      when Array
+        content.flat_map { format_content(_1) }
+      when URI
+        [{type: :image, source: {type: "url", url: content.to_s}}]
+      when LLM::File
+        if content.image?
+          [{type: :image, source: {type: "base64", media_type: content.mime_type, data: content.to_b64}}]
+        else
+          raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                          "is not an image, and therefore not supported by the " \
+                                          "Anthropic API"
+        end
+      when String
+        [{type: :text, text: content}]
+      when LLM::Message
+        format_content(content.content)
       else
-        content
+        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
+                                       "is not supported by the Anthropic API"
       end
     end
   end

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 +46,28 @@ 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: "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 +75,11 @@ 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/files.rb

@@ -83,7 +83,7 @@ class LLM::Gemini
       req["X-Goog-Upload-Offset"] = 0
       req["X-Goog-Upload-Command"] = "upload, finalize"
       file.with_io do |io|
-        req.body_stream = io
+        set_body_stream(req, io)
         res = request(http, req)
         LLM::Response::File.new(res)
       end
@@ -155,7 +155,7 @@ class LLM::Gemini
       @provider.instance_variable_get(:@secret)
     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/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
@@ -70,7 +70,7 @@ class LLM::Gemini
         contents: [{parts: [{text: prompt}, format_content(image)]}],
         generationConfig: {responseModalities: ["TEXT", "IMAGE"]}
       }.merge!(params)).b
-      req.body_stream = StringIO.new(body)
+      set_body_stream(req, StringIO.new(body))
       res = request(http, req)
       LLM::Response::Image.new(res).extend(response_parser)
     end
@@ -92,7 +92,7 @@ class LLM::Gemini
       @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