llm.rb 0.3.1 → 0.3.3

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "0.3.1"
+  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

@@ -48,7 +48,7 @@ RSpec.describe "LLM::OpenAI: completions" do
       openai.complete "What is your name? What age are you?", :user, messages: [
         {role: "system", content: "Answer all of my questions"},
         {role: "system", content: "Answer in the format: My name is <name> and I am <age> years old"},
-        {role: "system", content: "Your name is Pablo and you are 25 years old"},
+        {role: "system", content: "Your name is Pablo and you are 25 years old"}
       ]
     end
 
@@ -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) }
@@ -81,7 +100,7 @@ RSpec.describe "LLM::OpenAI: completions" do
 
   context "when given an unauthorized response",
           vcr: {cassette_name: "openai/completions/unauthorized_response"} do
-    subject(:response) { openai.complete(LLM::Message.new("Hello!", :user)) }
+    subject(:response) { openai.complete(LLM::Message.new(:user, "Hello!")) }
     let(:token) { "BADTOKEN" }
 
     it "raises an error" do

data/spec/openai/files_spec.rb

@@ -94,16 +94,16 @@ RSpec.describe "LLM::OpenAI::Files" do
         provider.files.create(file: LLM::File("spec/fixtures/documents/haiku2.txt"))
       ]
     end
-    subject(:file) { provider.files.all }
+    subject(:filelist) { provider.files.all }
 
     it "is successful" do
-      expect(file).to be_instance_of(LLM::Response::FileList)
+      expect(filelist).to be_instance_of(LLM::Response::FileList)
     ensure
       files.each { |file| provider.files.delete(file:) }
     end
 
     it "returns an array of file objects" do
-      expect(file).to match_array(
+      expect(filelist.files[0..1]).to match_array(
         [
           have_attributes(
             id: instance_of(String),

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

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm.rb
 version: !ruby/object:Gem::Version
-  version: 0.3.1
+  version: 0.3.3
 platform: ruby
 authors:
 - Antar Azri
@@ -9,7 +9,7 @@ authors:
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2025-04-26 00:00:00.000000000 Z
+date: 2025-04-29 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: webmock
@@ -162,6 +162,7 @@ files:
 - lib/llm/providers/anthropic.rb
 - lib/llm/providers/anthropic/error_handler.rb
 - lib/llm/providers/anthropic/format.rb
+- lib/llm/providers/anthropic/models.rb
 - lib/llm/providers/anthropic/response_parser.rb
 - lib/llm/providers/gemini.rb
 - lib/llm/providers/gemini/audio.rb
@@ -169,10 +170,12 @@ files:
 - lib/llm/providers/gemini/files.rb
 - lib/llm/providers/gemini/format.rb
 - lib/llm/providers/gemini/images.rb
+- lib/llm/providers/gemini/models.rb
 - lib/llm/providers/gemini/response_parser.rb
 - lib/llm/providers/ollama.rb
 - lib/llm/providers/ollama/error_handler.rb
 - lib/llm/providers/ollama/format.rb
+- lib/llm/providers/ollama/models.rb
 - lib/llm/providers/ollama/response_parser.rb
 - lib/llm/providers/openai.rb
 - lib/llm/providers/openai/audio.rb
@@ -180,6 +183,7 @@ files:
 - lib/llm/providers/openai/files.rb
 - lib/llm/providers/openai/format.rb
 - lib/llm/providers/openai/images.rb
+- lib/llm/providers/openai/models.rb
 - lib/llm/providers/openai/response_parser.rb
 - lib/llm/providers/openai/responses.rb
 - lib/llm/providers/voyageai.rb
@@ -195,30 +199,31 @@ files:
 - lib/llm/response/file.rb
 - lib/llm/response/filelist.rb
 - lib/llm/response/image.rb
+- lib/llm/response/modellist.rb
 - lib/llm/response/output.rb
 - lib/llm/utils.rb
 - lib/llm/version.rb
 - llm.gemspec
-- share/llm/models/anthropic.yml
-- share/llm/models/gemini.yml
-- share/llm/models/ollama.yml
-- share/llm/models/openai.yml
 - spec/anthropic/completion_spec.rb
 - spec/anthropic/embedding_spec.rb
+- spec/anthropic/models_spec.rb
 - spec/gemini/completion_spec.rb
 - spec/gemini/conversation_spec.rb
 - spec/gemini/embedding_spec.rb
 - spec/gemini/files_spec.rb
 - spec/gemini/images_spec.rb
+- spec/gemini/models_spec.rb
 - spec/llm/conversation_spec.rb
 - spec/ollama/completion_spec.rb
 - spec/ollama/conversation_spec.rb
 - spec/ollama/embedding_spec.rb
+- spec/ollama/models_spec.rb
 - spec/openai/audio_spec.rb
 - spec/openai/completion_spec.rb
 - spec/openai/embedding_spec.rb
 - spec/openai/files_spec.rb
 - spec/openai/images_spec.rb
+- spec/openai/models_spec.rb
 - spec/openai/responses_spec.rb
 - spec/readme_spec.rb
 - spec/setup.rb

data/share/llm/models/anthropic.yml

@@ -1,35 +0,0 @@
-claude-3-7-sonnet-20250219:
-  name: Claude 3.7 Sonnet
-  parameters: Unknown
-  description: Most intelligent Claude model with extended thinking and high capability
-  to_param: claude-3-7-sonnet-20250219
-
-claude-3-5-sonnet-20241022:
-  name: Claude 3.5 Sonnet (v2)
-  parameters: Unknown
-  description: High intelligence and capability; upgraded from previous Sonnet
-  to_param: claude-3-5-sonnet-20241022
-
-claude-3-5-sonnet-20240620:
-  name: Claude 3.5 Sonnet
-  parameters: Unknown
-  description: Intelligent and capable general-purpose model
-  to_param: claude-3-5-sonnet-20240620
-
-claude-3-5-haiku-20241022:
-  name: Claude 3.5 Haiku
-  parameters: Unknown
-  description: Blazing fast model for low-latency text generation
-  to_param: claude-3-5-haiku-20241022
-
-claude-3-opus-20240229:
-  name: Claude 3 Opus
-  parameters: Unknown
-  description: Top-level intelligence, fluency, and reasoning for complex tasks
-  to_param: claude-3-opus-20240229
-
-claude-3-haiku-20240307:
-  name: Claude 3 Haiku
-  parameters: Unknown
-  description: Fastest and most compact Claude model for near-instant responsiveness
-  to_param: claude-3-haiku-20240307

data/share/llm/models/gemini.yml

@@ -1,35 +0,0 @@
-gemini-2.5-pro-exp-03-25:
-  name: Gemini
-  parameters: Unknown
-  description: Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more
-  to_param: gemini-2.5-pro-exp-03-25
-
-gemini-2.0-flash:
-  name: Gemini
-  parameters: Unknown
-  description: Next generation features, speed, thinking, realtime streaming, and multimodal generation
-  to_param: gemini-2.0-flash
-
-gemini-2.0-flash-lite:
-  name: Gemini
-  parameters: Unknown
-  description: Cost efficiency and low latency
-  to_param: gemini-2.0-flash-lite
-
-gemini-1.5-flash:
-  name: Gemini
-  parameters: Unknown
-  description: Fast and versatile performance across a diverse variety of tasks
-  to_param: gemini-1.5-flash
-
-gemini-1.5-flash-8b:
-  name: Gemini
-  parameters: 8B
-  description: High volume and lower intelligence tasks
-  to_param: gemini-1.5-flash-8b
-
-gemini-1.5-pro:
-  name: Gemini
-  parameters: Unknown
-  description: Complex reasoning tasks requiring more intelligence
-  to_param: gemini-1.5-pro

data/share/llm/models/ollama.yml

@@ -1,155 +0,0 @@
----
-gemma3:1b:
-  name: Gemma
-  parameters: 1B
-  description: Lightweight version of Google's Gemma 3 language model, suitable for
-    low-resource environments
-  to_param: gemma3:1b
-gemma3:
-  name: Gemma
-  parameters: 4B
-  description: Balanced Gemma 3 model providing good accuracy with reasonable size
-  to_param: gemma3
-gemma3:12b:
-  name: Gemma
-  parameters: 12B
-  description: Larger Gemma 3 model offering improved reasoning and generation abilities
-  to_param: gemma3:12b
-gemma3:27b:
-  name: Gemma
-  parameters: 27B
-  description: High-end Gemma 3 model focused on top-tier performance and accuracy
-  to_param: gemma3:27b
-
-qwq:
-  name: QwQ
-  parameters: 32B
-  description: Large-scale model with high parameter count for complex tasks and
-    high-quality generation
-  to_param: qwq
-
-deepseek-r1:
-  name: DeepSeek-R1
-  parameters: 7B
-  description: Compact DeepSeek model optimized for research and experimentation
-  to_param: deepseek-r1
-deepseek-r1:671b:
-  name: DeepSeek-R1
-  parameters: 671B
-  description: Massive-scale DeepSeek model focused on advanced AI reasoning and
-    capabilities
-  to_param: deepseek-r1:671b
-deepseek-coder:
-  name: DeepSeek-Coder
-  parameters: 1.3B
-  description: Lightweight code generation model trained on 2T tokens of code and natural language
-  to_param: deepseek-coder
-deepseek-coder:6.7b:
-  name: DeepSeek-Coder
-  parameters: 6.7B
-  description: Mid-sized DeepSeek-Coder model offering a strong balance between speed and capability for code-related tasks
-  to_param: deepseek-coder:6.7b
-deepseek-coder:33b:
-  name: DeepSeek-Coder
-  parameters: 33B
-  description: Large DeepSeek-Coder model with high performance for code generation, understanding, and multilingual coding tasks
-  to_param: deepseek-coder:33b
-
-llama3.3:
-  name: Llama
-  parameters: 70B
-  description: Latest large Llama model designed for high-end performance in reasoning
-    and language tasks
-  to_param: llama3.3
-llama3.2:
-  name: Llama
-  parameters: 3B
-  description: Small but capable version of Llama 3.2 for lightweight applications
-  to_param: llama3.2
-llama3.2:1b:
-  name: Llama
-  parameters: 1B
-  description: Tiny version of Llama 3.2, extremely lightweight and fast
-  to_param: llama3.2:1b
-llama3.2-vision:
-  name: Llama Vision
-  parameters: 11B
-  description: Multimodal Llama 3.2 model with vision capabilities (images + text)
-  to_param: llama3.2-vision
-llama3.2-vision:90b:
-  name: Llama Vision
-  parameters: 90B
-  description: Large-scale vision-capable Llama model for advanced multimodal tasks
-  to_param: llama3.2-vision:90b
-llama3.1:
-  name: Llama
-  parameters: 8B
-  description: General-purpose Llama model designed for good accuracy and performance
-    balance
-  to_param: llama3.1
-llama3.1:405b:
-  name: Llama
-  parameters: 405B
-  description: Extremely large-scale version of Llama 3.1, suitable for advanced tasks
-  to_param: llama3.1:405b
-
-phi4:
-  name: Phi
-  parameters: 14B
-  description: Phi 4 is known for compact size and competitive performance in general
-    tasks
-  to_param: phi4
-phi4-mini:
-  name: Phi Mini
-  parameters: 3.8B
-  description: Lightweight variant of Phi 4 ideal for quick inference on constrained systems
-  to_param: phi4-mini
-
-mistral:
-  name: Mistral
-  parameters: 7B
-  description: Popular and versatile open model for general language tasks
-  to_param: mistral
-
-moondream:
-  name: Moondream
-  parameters: 1.4B
-  description: Compact vision-enabled model with strong general performance
-  to_param: moondream
-
-neural-chat:
-  name: Neural Chat
-  parameters: 7B
-  description: Chat-focused model fine-tuned for natural conversations
-  to_param: neural-chat
-
-starling-lm:
-  name: Starling
-  parameters: 7B
-  description: Model focused on instruction-following and conversational performance
-  to_param: starling-lm
-
-codellama:
-  name: Code Llama
-  parameters: 7B
-  description: Llama model variant fine-tuned specifically for code understanding
-    and generation
-  to_param: codellama
-
-llama2-uncensored:
-  name: Llama 2 Uncensored
-  parameters: 7B
-  description: Unfiltered version of Llama 2 for unrestricted language modeling
-  to_param: llama2-uncensored
-
-llava:
-  name: LLaVA
-  parameters: 7B
-  description: Multimodal model combining vision and language understanding
-  to_param: llava
-
-granite3.2:
-  name: Granite
-  parameters: 8B
-  description: IBM’s Granite model for enterprise-grade language applications
-  to_param: granite3.2

data/share/llm/models/openai.yml

@@ -1,46 +0,0 @@
----
-o3-mini:
-  name: OpenAI o3-mini
-  parameters: Unknown
-  description: Fast, flexible, intelligent reasoning model
-  to_param: o3-mini
-o1:
-  name: OpenAI o1
-  parameters: Unknown
-  description: High-intelligence reasoning model
-  to_param: o1
-o1-mini:
-  name: OpenAI o1-mini
-  parameters: Unknown
-  description: Faster, more affordable reasoning model than o1
-  to_param: o1-mini
-o1-pro:
-  name: OpenAI o1-pro
-  parameters: Unknown
-  description: More compute than o1 for better responses
-  to_param: o1-pro
-gpt-4.5-preview:
-  name: GPT-4.5 Preview
-  parameters: Unknown
-  description: Largest and most capable GPT model
-  to_param: gpt-4.5-preview
-gpt-4o:
-  name: GPT-4o
-  parameters: Unknown
-  description: Fast, intelligent, flexible GPT model
-  to_param: gpt-4o
-gpt-4o-mini:
-  name: GPT-4o Mini
-  parameters: Mini
-  description: Fast, affordable small model for focused tasks
-  to_param: gpt-4o-mini
-gpt-4o-realtime-preview:
-  name: GPT-4o Realtime
-  parameters: Unknown
-  description: Realtime model for text and audio inputs/outputs
-  to_param: gpt-4o-realtime-preview
-gpt-3.5-turbo:
-  name: GPT-3.5 Turbo
-  parameters: Unknown
-  description: Legacy GPT model for cheaper chat and non-chat tasks
-  to_param: gpt-3.5-turbo