llm.rb 0.3.0 → 0.3.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9073b7495fb9bdad2deec1d2c086b6d3b554c5a440dd884108a2fa8d12f7c8a9
-  data.tar.gz: 514902fc97de61dc18df8c22d51d9e86472a62e1ffb0c4ce4394b0684cddbd8a
+  metadata.gz: 3c55653b476d2fe6fe9457c89bc430c698668312ce89660a1d69abd8adf338eb
+  data.tar.gz: fe7d456bbb739eb091e82351839baef4c64d1d108a2c4cd7de3eb1b478982631
 SHA512:
-  metadata.gz: 0d0c35fa38ed3481872e29131d15e03e5a4bf0ad8a96c42ba64a5f48ed32584973d39b53ca630c966d54b6700a83a44abb1f4224c1bb9c1ca7f9e7a2d953e1c3
-  data.tar.gz: 8889034558c56a2bc1ff5321cf0ca45d82ac83ac7122c741e859caed7d060b34b99824cc53d20a5add4949dd135cf65c383f8400e7c112a44110fc1d4e0d2f4d
+  metadata.gz: 8cd55bb28eb92fea745d8b11062b2442bf4b2de88ecfb0b7dc99cfefd293bd45113088dd13ccfe7e251d2e369459da700f15725bae51c3d31d4bf68e19953138
+  data.tar.gz: dab47021b94d00e51e7d0ca3f92e2966170b9fd8ce7138e0728d2be7fb83da03104ff93cd7c54b760acca62dd03adf16462069db9eb5c30185743c25259105aa

data/README.md

@@ -3,7 +3,9 @@
 llm.rb is a lightweight library that provides a common interface
 and set of functionality for multiple Large Language Models (LLMs). It
 is designed to be simple, flexible, and easy to use – and it has been
-implemented with no dependencies outside Ruby's standard library.
+implemented with zero dependencies outside Ruby's standard library. See the
+[philosophy](#philosophy) section for more information on the design principles
+behind llm.rb.
 
 ## Examples
 
@@ -24,6 +26,7 @@ llm = LLM.openai("yourapikey")
 llm = LLM.gemini("yourapikey")
 llm = LLM.anthropic("yourapikey")
 llm = LLM.ollama(nil)
+llm = LLM.voyageai("yourapikey")
 ```
 
 ### Conversations
@@ -110,7 +113,7 @@ bot.messages.each { print "[#{_1.role}] ", _1.content, "\n" }
 #### Speech
 
 Some but not all providers implement audio generation capabilities that
-can create text from speech, transcribe audio to text, or translate
+can create speech from text, transcribe audio to text, or translate
 audio to text (usually English). The following example uses the OpenAI provider
 to create an audio file from a text prompt. The audio is then moved to
 `${HOME}/hello.mp3` as the final step. As always, consult the provider's
@@ -120,8 +123,6 @@ 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")
@@ -149,8 +150,6 @@ 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(
@@ -178,9 +177,8 @@ 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(
@@ -193,7 +191,7 @@ print res.text, "\n" # => "Good morning."
 
 #### Create
 
-Some but all LLM providers implement image generation capabilities that
+Some but not all LLM providers implement image generation capabilities that
 can create new images from a prompt, or edit an existing image with a
 prompt. The following example uses the OpenAI provider to create an
 image of a dog on a rocket to the moon. The image is then moved to
@@ -282,6 +280,84 @@ res.urls.each.with_index do |url, index|
 end
 ```
 
+### Files
+
+#### Create
+
+Most LLM providers provide a Files API where you can upload files
+that can be referenced from a prompt and llm.rb has first-class support
+for this feature. The following example uses the OpenAI provider to describe
+the contents of a PDF file after it has been uploaded. The file (an instance
+of [LLM::Response::File](https://0x1eef.github.io/x/llm.rb/LLM/Response/File.html))
+is passed directly to the chat method, and generally any object a prompt supports
+can be given to the chat method.
+
+Please also see provider-specific documentation for more provider-specific
+examples and documentation
+(eg
+[LLM::Gemini::Files](https://0x1eef.github.io/x/llm.rb/LLM/Gemini/Files.html),
+[LLM::OpenAI::Files](https://0x1eef.github.io/x/llm.rb/LLM/OpenAI/Files.html)):
+
+```ruby
+#!/usr/bin/env ruby
+require "llm"
+
+llm = LLM.openai(ENV["KEY"])
+bot = LLM::Chat.new(llm).lazy
+file = llm.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" }
+
+##
+# [assistant] This file is about OpenBSD, a free and open-source Unix-like operating system
+#             based on the Berkeley Software Distribution (BSD). It is known for its
+#             emphasis on security, code correctness, and code simplicity. The file
+#             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
@@ -354,6 +430,20 @@ llm.rb can be installed via rubygems.org:
 
 	gem install llm.rb
 
+## Philosophy
+
+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.
+
 ## License
 
 [BSD Zero Clause](https://choosealicense.com/licenses/0bsd/)

data/lib/llm/error.rb

@@ -4,25 +4,30 @@ module LLM
   ##
   # The superclass of all LLM errors
   class Error < RuntimeError
-    def initialize
+    def initialize(...)
       block_given? ? yield(self) : nil
+      super
     end
 
     ##
     # The superclass of all HTTP protocol errors
-    class BadResponse < Error
+    class ResponseError < Error
       ##
       # @return [Net::HTTPResponse]
       #  Returns the response associated with an error
       attr_accessor :response
     end
 
+    ##
+    # When a prompt is given an object that's not understood
+    PromptError = Class.new(Error)
+
     ##
     # HTTPUnauthorized
-    Unauthorized = Class.new(BadResponse)
+    Unauthorized = Class.new(ResponseError)
 
     ##
     # HTTPTooManyRequests
-    RateLimit = Class.new(BadResponse)
+    RateLimit = Class.new(ResponseError)
   end
 end

data/lib/llm/file.rb

@@ -41,6 +41,23 @@ class LLM::File
   def to_b64
     [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
+  def with_io
+    io = File.open(path, "rb")
+    yield(io)
+  ensure
+    io.close
+  end
 end
 
 ##

data/lib/llm/message.rb

@@ -50,6 +50,13 @@ module LLM
     end
     alias_method :eql?, :==
 
+    ##
+    # Returns true when the message is from the LLM
+    # @return [Boolean]
+    def assistant?
+      role == "assistant" || role == "model"
+    end
+
     ##
     # Returns a string representation of the message
     # @return [String]

data/lib/llm/multipart.rb

@@ -45,7 +45,9 @@ class LLM::Multipart
   # Returns the multipart request body
   # @return [String]
   def body
-    [*parts, "--#{@boundary}--\r\n"].inject(&:<<)
+    io = StringIO.new("".b)
+    [*parts, StringIO.new("--#{@boundary}--\r\n".b)].each { IO.copy_stream(_1.tap(&:rewind), io) }
+    io.tap(&:rewind)
   end
 
   private
@@ -61,7 +63,7 @@ class LLM::Multipart
 
   def multipart_header(type:, locals:)
     if type == :file
-      str = "".b
+      str = StringIO.new("".b)
       str << "--#{locals[:boundary]}" \
              "\r\n" \
              "Content-Disposition: form-data; name=\"#{locals[:key]}\";" \
@@ -70,7 +72,7 @@ class LLM::Multipart
              "Content-Type: #{locals[:content_type]}" \
              "\r\n\r\n"
     elsif type == :data
-      str = "".b
+      str = StringIO.new("".b)
       str << "--#{locals[:boundary]}" \
              "\r\n" \
              "Content-Disposition: form-data; name=\"#{locals[:key]}\"" \
@@ -82,17 +84,17 @@ class LLM::Multipart
 
   def file_part(key, file, locals)
     locals = locals.merge(attributes(file))
-    multipart_header(type: :file, locals:).tap do
-      _1 << File.binread(file.path)
-      _1 << "\r\n"
+    multipart_header(type: :file, locals:).tap do |io|
+      IO.copy_stream(file.path, io)
+      io << "\r\n"
     end
   end
 
   def data_part(key, value, locals)
     locals = locals.merge(value:)
-    multipart_header(type: :data, locals:).tap do
-      _1 << value.to_s
-      _1 << "\r\n"
+    multipart_header(type: :data, locals:).tap do |io|
+      io << value.to_s
+      io << "\r\n"
     end
   end
 end

data/lib/llm/provider.rb

@@ -4,19 +4,9 @@
 # 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_relative "http_client"
-  include LLM::HTTPClient
+  require "net/http"
 
   ##
   # @param [String] secret
@@ -79,7 +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: nil, **params)
     raise NotImplementedError
   end
 
@@ -222,6 +212,45 @@ class LLM::Provider
     raise NotImplementedError
   end
 
+  ##
+  # Initiates a HTTP request
+  # @param [Net::HTTP] http
+  #  The HTTP object to use for the request
+  # @param [Net::HTTPRequest] req
+  #  The request to send
+  # @param [Proc] b
+  #  A block to yield the response to (optional)
+  # @return [Net::HTTPResponse]
+  #  The response from the server
+  # @raise [LLM::Error::Unauthorized]
+  #  When authentication fails
+  # @raise [LLM::Error::RateLimit]
+  #  When the rate limit is exceeded
+  # @raise [LLM::Error::ResponseError]
+  #  When any other unsuccessful status code is returned
+  # @raise [LLM::Error::PromptError]
+  #  When given an object a provider does not understand
+  # @raise [SystemCallError]
+  #  When there is a network error at the operating system level
+  def request(http, req, &b)
+    res = http.request(req, &b)
+    case res
+    when Net::HTTPOK then res
+    else error_handler.new(res).raise_error!
+    end
+  end
+
+  ##
+  # @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
+
   ##
   # @param [String] provider
   #  The name of the provider

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

@@ -27,7 +27,7 @@ class LLM::Anthropic
       when Net::HTTPTooManyRequests
         raise LLM::Error::RateLimit.new { _1.response = res }, "Too many requests"
       else
-        raise LLM::Error::BadResponse.new { _1.response = res }, "Unexpected response"
+        raise LLM::Error::ResponseError.new { _1.response = res }, "Unexpected response"
       end
     end
   end

data/lib/llm/providers/anthropic.rb

@@ -28,7 +28,7 @@ module LLM
     #  The embedding model to use
     # @param [Hash] params
     #  Other embedding parameters
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return (see LLM::Provider#embed)
     def embed(input, token:, model: "voyage-2", **params)
       llm = LLM.voyageai(token)
@@ -44,7 +44,7 @@ module LLM
     # @param max_tokens The maximum number of tokens to generate
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @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)

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

@@ -37,7 +37,7 @@ class LLM::Gemini
     # @param [LLM::File, LLM::Response::File] file The input audio
     # @param [String] model The model to use
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::AudioTranscription]
     def create_transcription(file:, model: "gemini-1.5-flash", **params)
       res = @provider.complete [
@@ -61,7 +61,7 @@ class LLM::Gemini
     # @param [LLM::File, LLM::Response::File] file The input audio
     # @param [String] model The model to use
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::AudioTranslation]
     def create_translation(file:, model: "gemini-1.5-flash", **params)
       res = @provider.complete [

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

@@ -27,12 +27,12 @@ class LLM::Gemini
         if reason == "API_KEY_INVALID"
           raise LLM::Error::Unauthorized.new { _1.response = res }, "Authentication error"
         else
-          raise LLM::Error::BadResponse.new { _1.response = res }, "Unexpected response"
+          raise LLM::Error::ResponseError.new { _1.response = res }, "Unexpected response"
         end
       when Net::HTTPTooManyRequests
         raise LLM::Error::RateLimit.new { _1.response = res }, "Too many requests"
       else
-        raise LLM::Error::BadResponse.new { _1.response = res }, "Unexpected response"
+        raise LLM::Error::ResponseError.new { _1.response = res }, "Unexpected response"
       end
     end
 

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

@@ -17,9 +17,9 @@ class LLM::Gemini
   #   #!/usr/bin/env ruby
   #   require "llm"
   #
-  #   llm  = LLM.gemini(ENV["KEY"])
-  #   file = llm.files.create file: LLM::File("/audio/haiku.mp3")
+  #   llm = LLM.gemini(ENV["KEY"])
   #   bot = LLM::Chat.new(llm).lazy
+  #   file = llm.files.create file: LLM::File("/audio/haiku.mp3")
   #   bot.chat(file)
   #   bot.chat("Describe the audio file I sent to you")
   #   bot.chat("The audio file is the first message I sent to you.")
@@ -28,9 +28,9 @@ class LLM::Gemini
   #   #!/usr/bin/env ruby
   #   require "llm"
   #
-  #   llm  = LLM.gemini(ENV["KEY"])
-  #   file = llm.files.create file: LLM::File("/audio/haiku.mp3")
+  #   llm = LLM.gemini(ENV["KEY"])
   #   bot = LLM::Chat.new(llm).lazy
+  #   file = llm.files.create file: LLM::File("/audio/haiku.mp3")
   #   bot.chat(["Describe the audio file I sent to you", file])
   #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Files
@@ -52,7 +52,7 @@ class LLM::Gemini
     #   end
     # @see https://ai.google.dev/gemini-api/docs/files Gemini docs
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::FileList]
     def all(**params)
       query = URI.encode_www_form(params.merge!(key: secret))
@@ -75,16 +75,18 @@ class LLM::Gemini
     # @see https://ai.google.dev/gemini-api/docs/files Gemini docs
     # @param [File] file The file
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::File]
     def create(file:, **params)
       req = Net::HTTP::Post.new(request_upload_url(file:), {})
       req["content-length"] = file.bytesize
       req["X-Goog-Upload-Offset"] = 0
       req["X-Goog-Upload-Command"] = "upload, finalize"
-      req.body = File.binread(file.path)
-      res = request(http, req)
-      LLM::Response::File.new(res)
+      file.with_io do |io|
+        set_body_stream(req, io)
+        res = request(http, req)
+        LLM::Response::File.new(res)
+      end
     end
 
     ##
@@ -96,7 +98,7 @@ class LLM::Gemini
     # @see https://ai.google.dev/gemini-api/docs/files Gemini docs
     # @param [#name, String] file The file to get
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::File]
     def get(file:, **params)
       file_id = file.respond_to?(:name) ? file.name : file.to_s
@@ -114,7 +116,7 @@ class LLM::Gemini
     # @see https://ai.google.dev/gemini-api/docs/files Gemini docs
     # @param [#name, String] file The file to delete
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return [LLM::Response::File]
     def delete(file:, **params)
       file_id = file.respond_to?(:name) ? file.name : file.to_s
@@ -153,7 +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/images.rb

@@ -34,18 +34,19 @@ class LLM::Gemini
     # @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)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @note
     #  The prompt should make it clear you want to generate an image, or you
     #  might unexpectedly receive a purely textual response. This is due to how
     #  Gemini implements image generation under the hood.
     # @return [LLM::Response::Image]
     def create(prompt:, model: "gemini-2.0-flash-exp-image-generation", **params)
-      req = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{secret}", headers)
-      req.body = JSON.dump({
+      req  = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{secret}", headers)
+      body = JSON.dump({
         contents: [{parts: {text: prompt}}],
         generationConfig: {responseModalities: ["TEXT", "IMAGE"]}
       }.merge!(params))
+      req.body = body
       res = request(http, req)
       LLM::Response::Image.new(res).extend(response_parser)
     end
@@ -60,17 +61,16 @@ class LLM::Gemini
     # @param [LLM::File] image The image to edit
     # @param [String] prompt The prompt
     # @param [Hash] params Other parameters (see Gemini docs)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @note (see LLM::Gemini::Images#create)
     # @return [LLM::Response::Image]
     def edit(image:, prompt:, model: "gemini-2.0-flash-exp-image-generation", **params)
-      req = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{secret}", headers)
-      req.body = JSON.dump({
-        contents: [
-          {parts: [{text: prompt}, format_content(image)]}
-        ],
+      req  = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{secret}", headers)
+      body = JSON.dump({
+        contents: [{parts: [{text: prompt}, format_content(image)]}],
         generationConfig: {responseModalities: ["TEXT", "IMAGE"]}
-      }.merge!(params))
+      }.merge!(params)).b
+      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

data/lib/llm/providers/gemini.rb

@@ -49,7 +49,7 @@ module LLM
     # @param input (see LLM::Provider#embed)
     # @param model (see LLM::Provider#embed)
     # @param params (see LLM::Provider#embed)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return (see LLM::Provider#embed)
     def embed(input, model: "text-embedding-004", **params)
       path = ["/v1beta/models/#{model}", "embedContent?key=#{@secret}"].join(":")
@@ -67,14 +67,15 @@ module LLM
     # @param model (see LLM::Provider#complete)
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @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)
+      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)}).b
+      set_body_stream(req, StringIO.new(body))
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 

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

@@ -27,7 +27,7 @@ class LLM::Ollama
       when Net::HTTPTooManyRequests
         raise LLM::Error::RateLimit.new { _1.response = res }, "Too many requests"
       else
-        raise LLM::Error::BadResponse.new { _1.response = res }, "Unexpected response"
+        raise LLM::Error::ResponseError.new { _1.response = res }, "Unexpected response"
       end
     end
   end

data/lib/llm/providers/ollama.rb

@@ -37,7 +37,7 @@ module LLM
     # @param input (see LLM::Provider#embed)
     # @param model (see LLM::Provider#embed)
     # @param params (see LLM::Provider#embed)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return (see LLM::Provider#embed)
     def embed(input, model: "llama3.2", **params)
       params   = {model:}.merge!(params)
@@ -55,7 +55,7 @@ module LLM
     # @param model (see LLM::Provider#complete)
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
-    # @raise (see LLM::HTTPClient#request)
+    # @raise (see LLM::Provider#request)
     # @return (see LLM::Provider#complete)
     def complete(prompt, role = :user, model: "llama3.2", **params)
       params   = {model:, stream: false}.merge!(params)