llm.rb 0.1.0 → 0.2.1

data/lib/llm/providers/anthropic.rb

@@ -11,7 +11,6 @@ module LLM
     include Format
 
     HOST = "api.anthropic.com"
-    DEFAULT_PARAMS = {max_tokens: 1024, model: "claude-3-5-sonnet-20240620"}.freeze
 
     ##
     # @param secret (see LLM::Provider#initialize)
@@ -20,14 +19,17 @@ module LLM
     end
 
     ##
+    # Provides an embedding via VoyageAI per
+    # [Anthropic's recommendation](https://docs.anthropic.com/en/docs/build-with-claude/embeddings)
     # @param input (see LLM::Provider#embed)
+    # @param [String] token
+    #  Valid token for the VoyageAI API
+    # @param [Hash] params
+    #  Additional parameters to pass to the API
     # @return (see LLM::Provider#embed)
-    def embed(input, **params)
-      req = Net::HTTP::Post.new ["api.voyageai.com/v1", "embeddings"].join("/")
-      body = {input:, model: "voyage-2"}.merge!(params)
-      req = preflight(req, body)
-      res = request(@http, req)
-      Response::Embedding.new(res).extend(response_parser)
+    def embed(input, token:, **params)
+      llm = LLM.voyageai(token)
+      llm.embed(input, **params)
     end
 
     ##
@@ -36,20 +38,34 @@ module LLM
     # @param role (see LLM::Provider#complete)
     # @return (see LLM::Provider#complete)
     def complete(prompt, role = :user, **params)
-      req = Net::HTTP::Post.new ["/v1", "messages"].join("/")
+      params   = {max_tokens: 1024, model: "claude-3-5-sonnet-20240620"}.merge!(params)
+      req      = Net::HTTP::Post.new("/v1/messages", headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      params = DEFAULT_PARAMS.merge(params)
-      body = {messages: format(messages)}.merge!(params)
-      req = preflight(req, body)
-      res = request(@http, req)
+      req.body = JSON.dump({messages: format(messages)}.merge!(params))
+      res      = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # @return (see LLM::Provider#assistant_role)
+    def assistant_role
+      "assistant"
+    end
+
+    ##
+    # @return (see LLM::Provider#models)
+    def models
+      @models ||= load_models!("anthropic")
+    end
+
     private
 
-    def auth(req)
-      req["anthropic-version"] = "2023-06-01"
-      req["x-api-key"] = @secret
+    def headers
+      {
+        "Content-Type" => "application/json",
+        "x-api-key" => @secret,
+        "anthropic-version" => "2023-06-01"
+      }
     end
 
     def response_parser

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

@@ -7,7 +7,13 @@ class LLM::Gemini
     #  The messages to format
     # @return [Array<Hash>]
     def format(messages)
-      messages.map { {role: _1.role, parts: [format_content(_1.content)]} }
+      messages.map do
+        if Hash === _1
+          {role: _1[:role], parts: [format_content(_1[:content])]}
+        else
+          {role: _1.role, parts: [format_content(_1.content)]}
+        end
+      end
     end
 
     private

data/lib/llm/providers/gemini.rb

@@ -11,7 +11,6 @@ module LLM
     include Format
 
     HOST = "generativelanguage.googleapis.com"
-    DEFAULT_PARAMS = {model: "gemini-1.5-flash"}.freeze
 
     ##
     # @param secret (see LLM::Provider#initialize)
@@ -23,11 +22,10 @@ module LLM
     # @param input (see LLM::Provider#embed)
     # @return (see LLM::Provider#embed)
     def embed(input, **params)
-      path = ["/v1beta/models", "text-embedding-004"].join("/")
-      req = Net::HTTP::Post.new [path, "embedContent"].join(":")
-      body = {content: {parts: [{text: input}]}}
-      req = preflight(req, body)
-      res = request @http, req
+      path = ["/v1beta/models/text-embedding-004", "embedContent?key=#{@secret}"].join(":")
+      req = Net::HTTP::Post.new(path, headers)
+      req.body = JSON.dump({content: {parts: [{text: input}]}})
+      res = request(@http, req)
       Response::Embedding.new(res).extend(response_parser)
     end
 
@@ -37,20 +35,33 @@ module LLM
     # @param role (see LLM::Provider#complete)
     # @return (see LLM::Provider#complete)
     def complete(prompt, role = :user, **params)
-      params = DEFAULT_PARAMS.merge(params)
-      path = ["/v1beta/models", params.delete(:model)].join("/")
-      req = Net::HTTP::Post.new [path, "generateContent"].join(":")
+      params   = {model: "gemini-1.5-flash"}.merge!(params)
+      path     = ["/v1beta/models/#{params.delete(:model)}", "generateContent?key=#{@secret}"].join(":")
+      req      = Net::HTTP::Post.new(path, headers)
       messages = [*(params.delete(:messages) || []), LLM::Message.new(role, prompt)]
-      body = {contents: format(messages)}
-      req = preflight(req, body)
-      res = request(@http, req)
+      req.body = JSON.dump({contents: format(messages)})
+      res      = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # @return (see LLM::Provider#assistant_role)
+    def assistant_role
+      "model"
+    end
+
+    ##
+    # @return (see LLM::Provider#models)
+    def models
+      @models ||= load_models!("gemini")
+    end
+
     private
 
-    def auth(req)
-      req.path.replace [req.path, URI.encode_www_form(key: @secret)].join("?")
+    def headers
+      {
+        "Content-Type" => "application/json"
+      }
     end
 
     def response_parser

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

@@ -7,7 +7,13 @@ class LLM::Ollama
     #  The messages to format
     # @return [Array<Hash>]
     def format(messages)
-      messages.map { {role: _1.role, content: format_content(_1.content)} }
+      messages.map do
+        if Hash === _1
+          {role: _1[:role], content: format_content(_1[:content])}
+        else
+          {role: _1.role, content: format_content(_1.content)}
+        end
+      end
     end
 
     private

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

@@ -2,6 +2,19 @@
 
 class LLM::Ollama
   module ResponseParser
+    ##
+    # @param [Hash] body
+    #  The response body from the LLM provider
+    # @return [Hash]
+    def parse_embedding(body)
+      {
+        model: body["model"],
+        embeddings: body["data"].map { _1["embedding"] },
+        prompt_tokens: body.dig("usage", "prompt_tokens"),
+        total_tokens: body.dig("usage", "total_tokens")
+      }
+    end
+
     ##
     # @param [Hash] body
     #  The response body from the LLM provider

data/lib/llm/providers/ollama.rb

@@ -11,7 +11,6 @@ module LLM
     include Format
 
     HOST = "localhost"
-    DEFAULT_PARAMS = {model: "llama3.2", stream: false}.freeze
 
     ##
     # @param secret (see LLM::Provider#initialize)
@@ -19,25 +18,50 @@ module LLM
       super(secret, host: HOST, port: 11434, ssl: false, **)
     end
 
+    ##
+    # @param input (see LLM::Provider#embed)
+    # @return (see LLM::Provider#embed)
+    def embed(input, **params)
+      params   = {model: "llama3.2"}.merge!(params)
+      req      = Net::HTTP::Post.new("/v1/embeddings", headers)
+      req.body = JSON.dump({input:}.merge!(params))
+      res      = request(@http, req)
+      Response::Embedding.new(res).extend(response_parser)
+    end
+
     ##
     # @see https://github.com/ollama/ollama/blob/main/docs/api.md#generate-a-chat-completion Ollama docs
     # @param prompt (see LLM::Provider#complete)
     # @param role (see LLM::Provider#complete)
     # @return (see LLM::Provider#complete)
     def complete(prompt, role = :user, **params)
-      req = Net::HTTP::Post.new ["/api", "chat"].join("/")
+      params   = {model: "llama3.2", stream: false}.merge!(params)
+      req      = Net::HTTP::Post.new("/api/chat", headers)
       messages = [*(params.delete(:messages) || []), LLM::Message.new(role, prompt)]
-      params = DEFAULT_PARAMS.merge(params)
-      body = {messages: messages.map(&:to_h)}.merge!(params)
-      req = preflight(req, body)
-      res = request(@http, req)
+      req.body = JSON.dump({messages: messages.map(&:to_h)}.merge!(params))
+      res      = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # @return (see LLM::Provider#assistant_role)
+    def assistant_role
+      "assistant"
+    end
+
+    ##
+    # @return (see LLM::Provider#models)
+    def models
+      @models ||= load_models!("ollama")
+    end
+
     private
 
-    def auth(req)
-      req["Authorization"] = "Bearer #{@secret}"
+    def headers
+      {
+        "Content-Type" => "application/json",
+        "Authorization" => "Bearer #{@secret}"
+      }
     end
 
     def response_parser

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

@@ -7,7 +7,13 @@ class LLM::OpenAI
     #  The messages to format
     # @return [Array<Hash>]
     def format(messages)
-      messages.map { {role: _1.role, content: format_content(_1.content)} }
+      messages.map do
+        if Hash === _1
+          {role: _1[:role], content: format_content(_1[:content])}
+        else
+          {role: _1.role, content: format_content(_1.content)}
+        end
+      end
     end
 
     private

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

@@ -2,12 +2,14 @@
 
 class LLM::OpenAI
   module ResponseParser
+    ##
+    # @param [Hash] body
+    #  The response body from the LLM provider
+    # @return [Hash]
     def parse_embedding(body)
       {
         model: body["model"],
-        embeddings: body.dig("data").map do |data|
-          data["embedding"]
-        end,
+        embeddings: body["data"].map { _1["embedding"] },
         prompt_tokens: body.dig("usage", "prompt_tokens"),
         total_tokens: body.dig("usage", "total_tokens")
       }

data/lib/llm/providers/openai.rb

@@ -11,7 +11,6 @@ module LLM
     include Format
 
     HOST = "api.openai.com"
-    DEFAULT_PARAMS = {model: "gpt-4o-mini"}.freeze
 
     ##
     # @param secret (see LLM::Provider#initialize)
@@ -23,10 +22,9 @@ module LLM
     # @param input (see LLM::Provider#embed)
     # @return (see LLM::Provider#embed)
     def embed(input, **params)
-      req = Net::HTTP::Post.new ["/v1", "embeddings"].join("/")
-      body = {input:, model: "text-embedding-3-small"}.merge!(params)
-      req = preflight(req, body)
-      res = request @http, req
+      req = Net::HTTP::Post.new("/v1/embeddings", headers)
+      req.body = JSON.dump({input:, model: "text-embedding-3-small"}.merge!(params))
+      res = request(@http, req)
       Response::Embedding.new(res).extend(response_parser)
     end
 
@@ -36,19 +34,31 @@ module LLM
     # @param role (see LLM::Provider#complete)
     # @return (see LLM::Provider#complete)
     def complete(prompt, role = :user, **params)
-      req = Net::HTTP::Post.new ["/v1", "chat", "completions"].join("/")
+      params   = {model: "gpt-4o-mini"}.merge!(params)
+      req      = Net::HTTP::Post.new("/v1/chat/completions", headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      params = DEFAULT_PARAMS.merge(params)
-      body = {messages: format(messages)}.merge!(params)
-      req = preflight(req, body)
-      res = request(@http, req)
+      req.body = JSON.dump({messages: format(messages)}.merge!(params))
+      res      = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
+    ##
+    # @return (see LLM::Provider#assistant_role)
+    def assistant_role
+      "assistant"
+    end
+
+    def models
+      @models ||= load_models!("openai")
+    end
+
     private
 
-    def auth(req)
-      req["Authorization"] = "Bearer #{@secret}"
+    def headers
+      {
+        "Content-Type" => "application/json",
+        "Authorization" => "Bearer #{@secret}"
+      }
     end
 
     def response_parser

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

@@ -0,0 +1,32 @@
+# frozen_string_literal: true
+
+class LLM::VoyageAI
+  class ErrorHandler
+    ##
+    # @return [Net::HTTPResponse]
+    #  Non-2XX response from the server
+    attr_reader :res
+
+    ##
+    # @param [Net::HTTPResponse] res
+    #  The response from the server
+    # @return [LLM::OpenAI::ErrorHandler]
+    def initialize(res)
+      @res = res
+    end
+
+    ##
+    # @raise [LLM::Error]
+    #  Raises a subclass of {LLM::Error LLM::Error}
+    def raise_error!
+      case res
+      when Net::HTTPUnauthorized
+        raise LLM::Error::Unauthorized.new { _1.response = res }, "Authentication error"
+      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"
+      end
+    end
+  end
+end

data/lib/llm/providers/voyageai/response_parser.rb

@@ -0,0 +1,13 @@
+# frozen_string_literal: true
+
+class LLM::VoyageAI
+  module ResponseParser
+    def parse_embedding(body)
+      {
+        model: body["model"],
+        embeddings: body["data"].map { _1["embedding"] },
+        total_tokens: body.dig("usage", "total_tokens")
+      }
+    end
+  end
+end

data/lib/llm/providers/voyageai.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+module LLM
+  class VoyageAI < Provider
+    require_relative "voyageai/error_handler"
+    require_relative "voyageai/response_parser"
+    HOST = "api.voyageai.com"
+
+    ##
+    # @param secret (see LLM::Provider#initialize)
+    def initialize(secret, **)
+      super(secret, host: HOST, **)
+    end
+
+    ##
+    # Provides an embedding via VoyageAI per
+    # [Anthropic's recommendation](https://docs.anthropic.com/en/docs/build-with-claude/embeddings)
+    # @param input (see LLM::Provider#embed)
+    # @return (see LLM::Provider#embed)
+    def embed(input, **params)
+      req = Net::HTTP::Post.new("/v1/embeddings", headers)
+      req.body = JSON.dump({input:, model: "voyage-2"}.merge!(params))
+      res = request(@http, req)
+      Response::Embedding.new(res).extend(response_parser)
+    end
+
+    private
+
+    def headers
+      {
+        "Content-Type" => "application/json",
+        "Authorization" => "Bearer #{@secret}"
+      }
+    end
+
+    def response_parser
+      LLM::VoyageAI::ResponseParser
+    end
+
+    def error_handler
+      LLM::VoyageAI::ErrorHandler
+    end
+  end
+end

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "0.1.0"
+  VERSION = "0.2.1"
 end

data/lib/llm.rb

@@ -6,9 +6,10 @@ module LLM
   require_relative "llm/message"
   require_relative "llm/response"
   require_relative "llm/file"
+  require_relative "llm/model"
   require_relative "llm/provider"
   require_relative "llm/conversation"
-  require_relative "llm/lazy_conversation"
+  require_relative "llm/message_queue"
   require_relative "llm/core_ext/ostruct"
 
   module_function
@@ -16,32 +17,41 @@ module LLM
   ##
   # @param secret (see LLM::Anthropic#initialize)
   # @return (see LLM::Anthropic#initialize)
-  def anthropic(secret, **)
+  def anthropic(secret, options = {})
     require_relative "llm/providers/anthropic" unless defined?(LLM::Anthropic)
-    LLM::Anthropic.new(secret, **)
+    require_relative "llm/providers/voyageai" unless defined?(LLM::VoyageAI)
+    LLM::Anthropic.new(secret, **options)
+  end
+
+  ##
+  # @param secret (see LLM::VoyageAI#initialize)
+  # @return (see LLM::VoyageAI#initialize)
+  def voyageai(secret, options = {})
+    require_relative "llm/providers/voyageai" unless defined?(LLM::VoyageAI)
+    LLM::VoyageAI.new(secret, **options)
   end
 
   ##
   # @param secret (see LLM::Gemini#initialize)
   # @return (see LLM::Gemini#initialize)
-  def gemini(secret, **)
+  def gemini(secret, options = {})
     require_relative "llm/providers/gemini" unless defined?(LLM::Gemini)
-    LLM::Gemini.new(secret, **)
+    LLM::Gemini.new(secret, **options)
   end
 
   ##
   # @param host (see LLM::Ollama#initialize)
   # @return (see LLM::Ollama#initialize)
-  def ollama(secret)
+  def ollama(secret, options = {})
     require_relative "llm/providers/ollama" unless defined?(LLM::Ollama)
-    LLM::Ollama.new(secret)
+    LLM::Ollama.new(secret, **options)
   end
 
   ##
   # @param secret (see LLM::OpenAI#initialize)
   # @return (see LLM::OpenAI#initialize)
-  def openai(secret, **)
+  def openai(secret, options = {})
     require_relative "llm/providers/openai" unless defined?(LLM::OpenAI)
-    LLM::OpenAI.new(secret, **)
+    LLM::OpenAI.new(secret, **options)
   end
 end

data/llm.gemspec

@@ -25,16 +25,21 @@ Gem::Specification.new do |spec|
     "README.md", "LICENSE.txt",
     "lib/*.rb", "lib/**/*.rb",
     "spec/*.rb", "spec/**/*.rb",
-    "llm.gemspec"
+    "share/llm/models/*.yml", "llm.gemspec"
   ]
   spec.require_paths = ["lib"]
 
   spec.add_runtime_dependency "net-http", "~> 0.6.0"
   spec.add_runtime_dependency "json"
+  spec.add_runtime_dependency "yaml"
 
   spec.add_development_dependency "webmock", "~> 3.24.0"
   spec.add_development_dependency "yard", "~> 0.9.37"
   spec.add_development_dependency "kramdown", "~> 2.4"
   spec.add_development_dependency "webrick", "~> 1.8"
   spec.add_development_dependency "test-cmd.rb", "~> 0.12.0"
+  spec.add_development_dependency "rake", "~> 13.0"
+  spec.add_development_dependency "rspec", "~> 3.0"
+  spec.add_development_dependency "standard", "~> 1.40"
+  spec.add_development_dependency "vcr", "~> 6.0"
 end

data/share/llm/models/anthropic.yml

@@ -0,0 +1,35 @@
+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

@@ -0,0 +1,35 @@
+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