llm.rb 3.1.0 → 4.0.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: fa682f0c6793298daeaac88092cb52f03652cbbbf28adfd6b62f94b8a263f3f3
-  data.tar.gz: 1fb08983372becef70d866bdc4ee79ee8d8bba55ace5d4be4637a69e91341747
+  metadata.gz: 70c5f60cafc446edf8d1be15367ca77eb89e467785c77fbc7f758c29e761e8db
+  data.tar.gz: db9ec411a0c441e471a19a98624e0973815867c28ab89f8bcebc108b4dc11b3b
 SHA512:
-  metadata.gz: 720e09be8b25a9fde7d92887636d572edcdbd39a1b3a23ae1f44baaddb9f881c95927f63f16248f3a3d22da1704973f69f51309c487e1c97195175b772499b0d
-  data.tar.gz: 8cf35f7829b4e66ef002652643779658cf9c8cf8726f8b563eb5ca59ebcfc3a71eeb9b4cc473dfc4556324448855b6733fe3d48a73fb6e70fb91102544eb7061
+  metadata.gz: f06f9c0367ad3d3428ce7c5046aebd37e4bfea9eac483b5c08a448bac58e9c205d3566a246d3022e9bd6d1669a9f9b5244a475262b24303d845464f7ec3ce4de
+  data.tar.gz: 45e86e63614eb9f5c96111f6ba11d9ec3aae89572dd3959987ca4d0a190cd2d6f5c8ab8ad516cdf68512ec1c1285117616493b3670b4538564c097ec1aaa0ede

data/lib/llm/error.rb

@@ -35,10 +35,6 @@ module LLM
   # HTTPServerError
   ServerError = Class.new(Error)
 
-  ##
-  # When no images are found in a response
-  NoImageError = Class.new(Error)
-
   ##
   # When an given an input object that is not understood
   FormatError = Class.new(Error)

data/lib/llm/provider.rb

@@ -45,7 +45,7 @@ class LLM::Provider
   # @return [String]
   # @note The secret key is redacted in inspect for security reasons
   def inspect
-    "#<#{self.class.name}:0x#{object_id.to_s(16)} @key=[REDACTED] @http=#{@http.inspect}>"
+    "#<#{self.class.name}:0x#{object_id.to_s(16)} @key=[REDACTED] @client=#{@client.inspect}>"
   end
 
   ##

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

@@ -3,14 +3,12 @@
 class LLM::Gemini
   ##
   # The {LLM::Gemini::Images LLM::Gemini::Images} class provides an images
-  # object for interacting with [Gemini's images API](https://ai.google.dev/gemini-api/docs/image-generation).
-  # Please note that unlike OpenAI, which can return either URLs or base64-encoded strings,
-  # Gemini's images API will always return an image as a base64 encoded string that
-  # can be decoded into binary.
+  # object for interacting with Google's Imagen text-to-image models via the
+  # Imagen API: https://ai.google.dev/gemini-api/docs/imagen
+  #
   # @example
   #   #!/usr/bin/env ruby
   #   require "llm"
-  #
   #   llm = LLM.gemini(key: ENV["KEY"])
   #   res = llm.images.create prompt: "A dog on a rocket to the moon"
   #   IO.copy_stream res.images[0], "rocket.png"
@@ -31,21 +29,30 @@ class LLM::Gemini
     #   llm = LLM.gemini(key: ENV["KEY"])
     #   res = llm.images.create prompt: "A dog on a rocket to the moon"
     #   IO.copy_stream res.images[0], "rocket.png"
-    # @see https://ai.google.dev/gemini-api/docs/image-generation Gemini docs
+    # @see https://ai.google.dev/gemini-api/docs/imagen Imagen docs
     # @param [String] prompt The prompt
-    # @param [Hash] params Other parameters (see Gemini docs)
+    # @param [Integer] n The number of images to generate
+    # @param [String] image_size The size of the image ("1K", "2K", etc.)
+    # @param [String] aspect_ratio The aspect ratio of the image ("1:1", "16:9", etc.)
+    # @param [String] person_generation Allow the model to generate images of people ("dont_allow", "allow_adult", "allow_all")
+    # @param [String] model The model to use
+    # @param [Hash] params Other parameters (see Imagen docs)
     # @raise (see LLM::Provider#request)
-    # @raise [LLM::NoImageError] when no images are returned
     # @return [LLM::Response]
-    def create(prompt:, model: "gemini-2.5-flash-image", **params)
-      req  = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{key}", headers)
+    def create(prompt:, n: 1, image_size: nil, aspect_ratio: nil, person_generation: nil, model: "imagen-4.0-generate-001", **params)
+      req  = Net::HTTP::Post.new("/v1beta/models/#{model}:predict?key=#{key}", headers)
       body = LLM.json.dump({
-        contents: [{parts: [{text: create_prompt}, {text: prompt}]}],
-        generationConfig: {responseModalities: ["TEXT", "IMAGE"]}
-      }.merge!(params))
+        parameters: {
+          sampleCount: n,
+          imageSize: image_size,
+          aspectRatio: aspect_ratio,
+          personGeneration: person_generation
+        }.compact.merge!(params),
+        instances: [{prompt:}]
+      })
       req.body = body
       res = execute(request: req)
-      validate ResponseAdapter.adapt(res, type: :image)
+      ResponseAdapter.adapt(res, type: :image)
     end
 
     ##
@@ -59,19 +66,10 @@ class LLM::Gemini
     # @param [String] prompt The prompt
     # @param [Hash] params Other parameters (see Gemini docs)
     # @raise (see LLM::Provider#request)
-    # @raise [LLM::NoImageError] when no images are returned
     # @note (see LLM::Gemini::Images#create)
     # @return [LLM::Response]
     def edit(image:, prompt:, model: "gemini-2.5-flash-image", **params)
-      req   = Net::HTTP::Post.new("/v1beta/models/#{model}:generateContent?key=#{key}", headers)
-      image = LLM::Object.from(value: LLM.File(image), kind: :local_file)
-      body  = LLM.json.dump({
-        contents: [{parts: [{text: edit_prompt}, {text: prompt}, adapter.adapt_content(image)]}],
-        generationConfig: {responseModalities: ["TEXT", "IMAGE"]}
-      }.merge!(params)).b
-      set_body_stream(req, StringIO.new(body))
-      res = execute(request: req)
-      validate ResponseAdapter.adapt(res, type: :image)
+      raise NotImplementedError, "image editing is not yet supported by Gemini"
     end
 
     ##
@@ -91,36 +89,6 @@ class LLM::Gemini
       @provider.instance_variable_get(:@key)
     end
 
-    def create_prompt
-      <<~PROMPT
-        ## Context
-        Your task is to generate one or more image(s) based on the user's instructions.
-        The user will provide you with text only.
-
-        ## Instructions
-        1. The model *MUST* generate image(s) based on the user text alone.
-        2. The model *MUST NOT* generate anything else.
-      PROMPT
-    end
-
-    def edit_prompt
-      <<~PROMPT
-        ## Context
-        Your task is to edit the provided image based on the user's instructions.
-        The user will provide you with both text and an image.
-
-        ## Instructions
-        1. The model *MUST* edit the provided image based on the user's instructions
-        2. The model *MUST NOT* generate a new image.
-        3. The model *MUST NOT* generate anything else.
-      PROMPT
-    end
-
-    def validate(res)
-      return res unless res.images.empty?
-      raise LLM::NoImageError.new { _1.response = res.res }, "no images found in response"
-    end
-
     [:headers, :execute, :set_body_stream].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end

data/lib/llm/providers/gemini/response_adapter/image.rb

@@ -5,13 +5,9 @@ module LLM::Gemini::ResponseAdapter
     ##
     # @return [Array<StringIO>]
     def images
-      candidates.flat_map do |candidate|
-        parts = candidate&.dig("content", "parts") || []
-        parts.filter_map do
-          data = _1.dig("inlineData", "data")
-          next unless data
-          StringIO.new(data.unpack1("m0"))
-        end
+      (body.predictions || []).map do
+        b64 = _1["bytesBase64Encoded"]
+        StringIO.new(b64.unpack1("m0"))
       end
     end
 
@@ -22,10 +18,5 @@ module LLM::Gemini::ResponseAdapter
     #  will always return an empty array.
     # @return [Array<String>]
     def urls = []
-
-    ##
-    # Returns one or more candidates, or an empty array
-    # @return [Array<Hash>]
-    def candidates = body.candidates || []
   end
 end

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

@@ -43,7 +43,7 @@ class LLM::Gemini
 
     def merge_candidates!(deltas)
       deltas.each do |delta|
-        index = delta["index"]
+        index = delta["index"].to_i
         @body["candidates"][index] ||= {"content" => {"parts" => []}}
         candidate = @body["candidates"][index]
         delta.each do |key, value|
@@ -81,6 +81,8 @@ class LLM::Gemini
           parts << delta
         elsif delta["fileData"]
           parts << delta
+        else
+          parts << delta
         end
       end
     end

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

@@ -41,7 +41,7 @@ class LLM::OpenAI
         index = choice["index"]
         if @body["choices"][index]
           target_message = @body["choices"][index]["message"]
-          delta = choice["delta"]
+          delta = choice["delta"] || {}
           delta.each do |key, value|
             if key == "content"
               target_message[key] ||= +""
@@ -56,7 +56,7 @@ class LLM::OpenAI
         else
           message_hash = {"role" => "assistant"}
           @body["choices"][index] = {"message" => message_hash}
-          choice["delta"].each do |key, value|
+          (choice["delta"] || {}).each do |key, value|
             if key == "content"
               @io << value if @io.respond_to?(:<<)
               message_hash[key] = value

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "3.1.0"
+  VERSION = "4.0.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm.rb
 version: !ruby/object:Gem::Version
-  version: 3.1.0
+  version: 4.0.0
 platform: ruby
 authors:
 - Antar Azri