llm.rb 2.1.0 → 3.0.0

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

@@ -13,7 +13,6 @@ class LLM::OpenAI
   #  chunks = llm.vector_stores.search(vector: store, query: "What is Ruby?")
   #  chunks.each { |chunk| puts chunk }
   class VectorStores
-    require_relative "response/enumerable"
     PollError = Class.new(LLM::Error)
 
     INTERVAL = 0.01
@@ -34,7 +33,7 @@ class LLM::OpenAI
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/vector_stores?#{query}", headers)
       res = execute(request: req)
-      LLM::Response.new(res).extend(LLM::OpenAI::Response::Enumerable)
+      ResponseAdapter.adapt(res, type: :enumerable)
     end
 
     ##
@@ -47,7 +46,7 @@ class LLM::OpenAI
     # @see https://platform.openai.com/docs/api-reference/vector_stores/create OpenAI docs
     def create(name:, file_ids: nil, **params)
       req = Net::HTTP::Post.new("/v1/vector_stores", headers)
-      req.body = JSON.dump(params.merge({name:, file_ids:}).compact)
+      req.body = LLM.json.dump(params.merge({name:, file_ids:}).compact)
       res = execute(request: req)
       LLM::Response.new(res)
     end
@@ -85,7 +84,7 @@ class LLM::OpenAI
     def modify(vector:, name: nil, **params)
       vector_id = vector.respond_to?(:id) ? vector.id : vector
       req = Net::HTTP::Post.new("/v1/vector_stores/#{vector_id}", headers)
-      req.body = JSON.dump(params.merge({name:}).compact)
+      req.body = LLM.json.dump(params.merge({name:}).compact)
       res = execute(request: req)
       LLM::Response.new(res)
     end
@@ -114,9 +113,9 @@ class LLM::OpenAI
     def search(vector:, query:, **params)
       vector_id = vector.respond_to?(:id) ? vector.id : vector
       req = Net::HTTP::Post.new("/v1/vector_stores/#{vector_id}/search", headers)
-      req.body = JSON.dump(params.merge({query:}).compact)
+      req.body = LLM.json.dump(params.merge({query:}).compact)
       res = execute(request: req)
-      LLM::Response.new(res).extend(LLM::OpenAI::Response::Enumerable)
+      ResponseAdapter.adapt(res, type: :enumerable)
     end
 
     ##
@@ -131,7 +130,7 @@ class LLM::OpenAI
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/vector_stores/#{vector_id}/files?#{query}", headers)
       res = execute(request: req)
-      LLM::Response.new(res).extend(LLM::OpenAI::Response::Enumerable)
+      ResponseAdapter.adapt(res, type: :enumerable)
     end
 
     ##
@@ -147,7 +146,7 @@ class LLM::OpenAI
       vector_id = vector.respond_to?(:id) ? vector.id : vector
       file_id = file.respond_to?(:id) ? file.id : file
       req = Net::HTTP::Post.new("/v1/vector_stores/#{vector_id}/files", headers)
-      req.body = JSON.dump(params.merge({file_id:, attributes:}).compact)
+      req.body = LLM.json.dump(params.merge({file_id:, attributes:}).compact)
       res = execute(request: req)
       LLM::Response.new(res)
     end
@@ -176,7 +175,7 @@ class LLM::OpenAI
       vector_id = vector.respond_to?(:id) ? vector.id : vector
       file_id = file.respond_to?(:id) ? file.id : file
       req = Net::HTTP::Post.new("/v1/vector_stores/#{vector_id}/files/#{file_id}", headers)
-      req.body = JSON.dump(params.merge({attributes:}).compact)
+      req.body = LLM.json.dump(params.merge({attributes:}).compact)
       res = execute(request: req)
       LLM::Response.new(res)
     end

data/lib/llm/providers/openai.rb

@@ -14,11 +14,9 @@ module LLM
   #   bot.chat ["Tell me about this photo", File.open("/images/capybara.jpg", "rb")]
   #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class OpenAI < Provider
-    require_relative "openai/response/embedding"
-    require_relative "openai/response/completion"
-    require_relative "openai/response/web_search"
     require_relative "openai/error_handler"
-    require_relative "openai/format"
+    require_relative "openai/request_adapter"
+    require_relative "openai/response_adapter"
     require_relative "openai/stream_parser"
     require_relative "openai/models"
     require_relative "openai/responses"
@@ -28,7 +26,7 @@ module LLM
     require_relative "openai/moderations"
     require_relative "openai/vector_stores"
 
-    include Format
+    include RequestAdapter
 
     HOST = "api.openai.com"
 
@@ -48,9 +46,9 @@ module LLM
     # @return (see LLM::Provider#embed)
     def embed(input, model: "text-embedding-3-small", **params)
       req = Net::HTTP::Post.new("/v1/embeddings", headers)
-      req.body = JSON.dump({input:, model:}.merge!(params))
+      req.body = LLM.json.dump({input:, model:}.merge!(params))
       res = execute(request: req)
-      LLM::Response.new(res).extend(LLM::OpenAI::Response::Embedding)
+      ResponseAdapter.adapt(res, type: :embedding)
     end
 
     ##
@@ -66,17 +64,16 @@ module LLM
     def complete(prompt, params = {})
       params = {role: :user, model: default_model}.merge!(params)
       tools  = resolve_tools(params.delete(:tools))
-      params = [params, format_schema(params), format_tools(tools)].inject({}, &:merge!).compact
+      params = [params, adapt_schema(params), adapt_tools(tools)].inject({}, &:merge!).compact
       role, stream = params.delete(:role), params.delete(:stream)
       params[:stream] = true if stream.respond_to?(:<<) || stream == true
       params[:stream_options] = {include_usage: true}.merge!(params[:stream_options] || {}) if params[:stream]
       req = Net::HTTP::Post.new(completions_path, headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      body = JSON.dump({messages: format(messages, :complete).flatten}.merge!(params))
+      body = LLM.json.dump({messages: adapt(messages, mode: :complete).flatten}.merge!(params))
       set_body_stream(req, StringIO.new(body))
       res = execute(request: req, stream:)
-      LLM::Response.new(res)
-        .extend(LLM::OpenAI::Response::Completion)
+      ResponseAdapter.adapt(res, type: :completion)
         .extend(Module.new { define_method(:__tools__) { tools } })
     end
 
@@ -177,9 +174,10 @@ module LLM
     # @param query [String] The search query.
     # @return [LLM::Response] The response from the LLM provider.
     def web_search(query:)
-      responses
-        .create(query, store: false, tools: [server_tools[:web_search]])
-        .extend(LLM::OpenAI::Response::WebSearch)
+      ResponseAdapter.adapt(
+        responses.create(query, store: false, tools: [server_tools[:web_search]]),
+        type: :web_search
+      )
     end
 
     private

data/lib/llm/response.rb

@@ -25,12 +25,9 @@ module LLM
 
     ##
     # Returns the response body
-    # @return [Hash, String]
+    # @return [LLM::Object, String]
     def body
-      @body ||= case @res["content-type"]
-      when %r|\Aapplication/json\s*| then LLM::Object.from_hash(JSON.parse(@res.body))
-      else @res.body
-      end
+      @res.body
     end
 
     ##

data/lib/llm/usage.rb

@@ -0,0 +1,10 @@
+# frozen_string_literal: true
+
+##
+# The {LLM::Usage LLM::Usage} class represents token usage for
+# a given conversation or completion. As a conversation grows,
+# so does the number of tokens used. This class helps track
+# the number of input, output, and total tokens. It can also help
+# track usage of the context window (which may vary by model).
+class LLM::Usage < Struct.new(:input_tokens, :output_tokens, :total_tokens, keyword_init: true)
+end

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "2.1.0"
+  VERSION = "3.0.0"
 end

data/lib/llm.rb

@@ -2,12 +2,15 @@
 
 module LLM
   require "stringio"
+  require_relative "llm/json_adapter"
+  require_relative "llm/error"
+  require_relative "llm/contract"
+  require_relative "llm/usage"
   require_relative "llm/builder"
   require_relative "llm/schema"
   require_relative "llm/object"
   require_relative "llm/version"
   require_relative "llm/utils"
-  require_relative "llm/error"
   require_relative "llm/message"
   require_relative "llm/response"
   require_relative "llm/mime"
@@ -28,6 +31,35 @@ module LLM
 
   module_function
 
+  ##
+  # Returns the JSON adapter used by the library
+  # @return [Class]
+  #  Returns a class that responds to +dump+ and +load+
+  def json
+    @json ||= JSONAdapter::JSON
+  end
+
+  ##
+  # Sets the JSON adapter used by the library
+  # @param [Class, String, Symbol] adapter
+  #  A JSON adapter class or its name
+  # @return [void]
+  def json=(adapter)
+    @json = case adapter.to_s
+    when "JSON", "json" then JSONAdapter::JSON
+    when "Oj", "oj" then JSONAdapter::Oj
+    when "Yajl", "yajl" then JSONAdapter::Yajl
+    else
+      is_class = Class === adapter
+      is_subclass = is_class && adapter.ancestors.include?(LLM::JSONAdapter)
+      if is_subclass
+        adapter
+      else
+        raise TypeError, "Adapter must be a subclass of LLM::JSONAdapter"
+      end
+    end
+  end
+
   ##
   # @param (see LLM::Provider#initialize)
   # @return (see LLM::Anthropic#initialize)

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm.rb
 version: !ruby/object:Gem::Version
-  version: 2.1.0
+  version: 3.0.0
 platform: ruby
 authors:
 - Antar Azri
@@ -182,6 +182,8 @@ files:
 - lib/llm/buffer.rb
 - lib/llm/builder.rb
 - lib/llm/client.rb
+- lib/llm/contract.rb
+- lib/llm/contract/completion.rb
 - lib/llm/error.rb
 - lib/llm/eventhandler.rb
 - lib/llm/eventstream.rb
@@ -189,9 +191,11 @@ files:
 - lib/llm/eventstream/parser.rb
 - lib/llm/file.rb
 - lib/llm/function.rb
+- lib/llm/json_adapter.rb
 - lib/llm/message.rb
 - lib/llm/mime.rb
 - lib/llm/multipart.rb
+- lib/llm/multipart/enumerator_io.rb
 - lib/llm/object.rb
 - lib/llm/object/builder.rb
 - lib/llm/object/kernel.rb
@@ -199,62 +203,66 @@ files:
 - lib/llm/providers/anthropic.rb
 - lib/llm/providers/anthropic/error_handler.rb
 - lib/llm/providers/anthropic/files.rb
-- lib/llm/providers/anthropic/format.rb
-- lib/llm/providers/anthropic/format/completion_format.rb
 - lib/llm/providers/anthropic/models.rb
-- lib/llm/providers/anthropic/response/completion.rb
-- lib/llm/providers/anthropic/response/enumerable.rb
-- lib/llm/providers/anthropic/response/file.rb
-- lib/llm/providers/anthropic/response/web_search.rb
+- lib/llm/providers/anthropic/request_adapter.rb
+- lib/llm/providers/anthropic/request_adapter/completion.rb
+- lib/llm/providers/anthropic/response_adapter.rb
+- lib/llm/providers/anthropic/response_adapter/completion.rb
+- lib/llm/providers/anthropic/response_adapter/enumerable.rb
+- lib/llm/providers/anthropic/response_adapter/file.rb
+- lib/llm/providers/anthropic/response_adapter/web_search.rb
 - lib/llm/providers/anthropic/stream_parser.rb
 - lib/llm/providers/deepseek.rb
-- lib/llm/providers/deepseek/format.rb
-- lib/llm/providers/deepseek/format/completion_format.rb
+- lib/llm/providers/deepseek/request_adapter.rb
+- lib/llm/providers/deepseek/request_adapter/completion.rb
 - lib/llm/providers/gemini.rb
 - lib/llm/providers/gemini/audio.rb
 - lib/llm/providers/gemini/error_handler.rb
 - lib/llm/providers/gemini/files.rb
-- lib/llm/providers/gemini/format.rb
-- lib/llm/providers/gemini/format/completion_format.rb
 - lib/llm/providers/gemini/images.rb
 - lib/llm/providers/gemini/models.rb
-- lib/llm/providers/gemini/response/completion.rb
-- lib/llm/providers/gemini/response/embedding.rb
-- lib/llm/providers/gemini/response/file.rb
-- lib/llm/providers/gemini/response/files.rb
-- lib/llm/providers/gemini/response/image.rb
-- lib/llm/providers/gemini/response/models.rb
-- lib/llm/providers/gemini/response/web_search.rb
+- lib/llm/providers/gemini/request_adapter.rb
+- lib/llm/providers/gemini/request_adapter/completion.rb
+- lib/llm/providers/gemini/response_adapter.rb
+- lib/llm/providers/gemini/response_adapter/completion.rb
+- lib/llm/providers/gemini/response_adapter/embedding.rb
+- lib/llm/providers/gemini/response_adapter/file.rb
+- lib/llm/providers/gemini/response_adapter/files.rb
+- lib/llm/providers/gemini/response_adapter/image.rb
+- lib/llm/providers/gemini/response_adapter/models.rb
+- lib/llm/providers/gemini/response_adapter/web_search.rb
 - lib/llm/providers/gemini/stream_parser.rb
 - lib/llm/providers/llamacpp.rb
 - lib/llm/providers/ollama.rb
 - lib/llm/providers/ollama/error_handler.rb
-- lib/llm/providers/ollama/format.rb
-- lib/llm/providers/ollama/format/completion_format.rb
 - lib/llm/providers/ollama/models.rb
-- lib/llm/providers/ollama/response/completion.rb
-- lib/llm/providers/ollama/response/embedding.rb
+- lib/llm/providers/ollama/request_adapter.rb
+- lib/llm/providers/ollama/request_adapter/completion.rb
+- lib/llm/providers/ollama/response_adapter.rb
+- lib/llm/providers/ollama/response_adapter/completion.rb
+- lib/llm/providers/ollama/response_adapter/embedding.rb
 - lib/llm/providers/ollama/stream_parser.rb
 - lib/llm/providers/openai.rb
 - lib/llm/providers/openai/audio.rb
 - lib/llm/providers/openai/error_handler.rb
 - lib/llm/providers/openai/files.rb
-- lib/llm/providers/openai/format.rb
-- lib/llm/providers/openai/format/completion_format.rb
-- lib/llm/providers/openai/format/moderation_format.rb
-- lib/llm/providers/openai/format/respond_format.rb
 - lib/llm/providers/openai/images.rb
 - lib/llm/providers/openai/models.rb
 - lib/llm/providers/openai/moderations.rb
-- lib/llm/providers/openai/response/audio.rb
-- lib/llm/providers/openai/response/completion.rb
-- lib/llm/providers/openai/response/embedding.rb
-- lib/llm/providers/openai/response/enumerable.rb
-- lib/llm/providers/openai/response/file.rb
-- lib/llm/providers/openai/response/image.rb
-- lib/llm/providers/openai/response/moderations.rb
-- lib/llm/providers/openai/response/responds.rb
-- lib/llm/providers/openai/response/web_search.rb
+- lib/llm/providers/openai/request_adapter.rb
+- lib/llm/providers/openai/request_adapter/completion.rb
+- lib/llm/providers/openai/request_adapter/moderation.rb
+- lib/llm/providers/openai/request_adapter/respond.rb
+- lib/llm/providers/openai/response_adapter.rb
+- lib/llm/providers/openai/response_adapter/audio.rb
+- lib/llm/providers/openai/response_adapter/completion.rb
+- lib/llm/providers/openai/response_adapter/embedding.rb
+- lib/llm/providers/openai/response_adapter/enumerable.rb
+- lib/llm/providers/openai/response_adapter/file.rb
+- lib/llm/providers/openai/response_adapter/image.rb
+- lib/llm/providers/openai/response_adapter/moderations.rb
+- lib/llm/providers/openai/response_adapter/responds.rb
+- lib/llm/providers/openai/response_adapter/web_search.rb
 - lib/llm/providers/openai/responses.rb
 - lib/llm/providers/openai/responses/stream_parser.rb
 - lib/llm/providers/openai/stream_parser.rb
@@ -276,6 +284,7 @@ files:
 - lib/llm/server_tool.rb
 - lib/llm/tool.rb
 - lib/llm/tool/param.rb
+- lib/llm/usage.rb
 - lib/llm/utils.rb
 - lib/llm/version.rb
 - llm.gemspec

data/lib/llm/providers/anthropic/response/completion.rb

@@ -1,39 +0,0 @@
-# frozen_string_literal: true
-
-module LLM::Anthropic::Response
-  module Completion
-    def choices = format_choices
-    def role = body.role
-    def model = body.model
-    def prompt_tokens = body.usage["input_tokens"] || 0
-    def completion_tokens = body.usage["output_tokens"] || 0
-    def total_tokens = prompt_tokens + completion_tokens
-
-    private
-
-    def format_choices
-      texts.map.with_index do |choice, index|
-        extra = {
-          index:, response: self,
-          tool_calls: format_tool_calls(tools), original_tool_calls: tools
-        }
-        LLM::Message.new(role, choice["text"], extra)
-      end
-    end
-
-    def format_tool_calls(tools)
-      (tools || []).filter_map do |tool|
-        tool = {
-          id: tool.id,
-          name: tool.name,
-          arguments: tool.input
-        }
-        LLM::Object.new(tool)
-      end
-    end
-
-    def parts = body.content
-    def texts = @texts ||= LLM::Object.from_hash(parts.select { _1["type"] == "text" })
-    def tools = @tools ||= LLM::Object.from_hash(parts.select { _1["type"] == "tool_use" })
-  end
-end

data/lib/llm/providers/gemini/response/completion.rb

@@ -1,35 +0,0 @@
-# frozen_string_literal: true
-
-module LLM::Gemini::Response
-  module Completion
-    def model = body.modelVersion
-    def prompt_tokens = body.usageMetadata.promptTokenCount
-    def completion_tokens = body.usageMetadata.candidatesTokenCount
-    def total_tokens = body.usageMetadata.totalTokenCount
-    def choices = format_choices
-
-    private
-
-    def format_choices
-      candidates.map.with_index do |choice, index|
-        choice = LLM::Object.from_hash(choice)
-        content = choice.content || LLM::Object.new
-        role = content.role || "model"
-        parts = content.parts || [{"text" => choice.finishReason}]
-        text  = parts.filter_map { _1["text"] }.join
-        tools = parts.filter_map { _1["functionCall"] }
-        extra = {index:, response: self, tool_calls: format_tool_calls(tools), original_tool_calls: tools}
-        LLM::Message.new(role, text, extra)
-      end
-    end
-
-    def format_tool_calls(tools)
-      (tools || []).map do |tool|
-        function = {name: tool.name, arguments: tool.args}
-        LLM::Object.new(function)
-      end
-    end
-
-    def candidates = body.candidates || []
-  end
-end

data/lib/llm/providers/ollama/response/completion.rb

@@ -1,28 +0,0 @@
-# frozen_string_literal: true
-
-module LLM::Ollama::Response
-  module Completion
-    def model = body.model
-    def prompt_tokens = body.prompt_eval_count || 0
-    def completion_tokens = body.eval_count || 0
-    def total_tokens = prompt_tokens + completion_tokens
-    def message = body.message
-    def choices = [format_choices]
-
-    private
-
-    def format_choices
-      role, content, calls = message.to_h.values_at("role", "content", "tool_calls")
-      extra = {response: self, tool_calls: format_tool_calls(calls)}
-      LLM::Message.new(role, content, extra)
-    end
-
-    def format_tool_calls(tools)
-      return [] unless tools
-      tools.filter_map do |tool|
-        next unless tool["function"]
-        LLM::Object.new(tool["function"])
-      end
-    end
-  end
-end

data/lib/llm/providers/openai/response/completion.rb

@@ -1,40 +0,0 @@
-# frozen_string_literal: true
-
-module LLM::OpenAI::Response
-  module Completion
-    def choices
-      body.choices.map.with_index do |choice, index|
-        choice = LLM::Object.from_hash(choice)
-        message = choice.message
-        extra = {
-          index:, response: self,
-          logprobs: choice.logprobs,
-          tool_calls: format_tool_calls(message.tool_calls),
-          original_tool_calls: message.tool_calls
-        }
-        LLM::Message.new(message.role, message.content, extra)
-      end
-    end
-    alias_method :messages, :choices
-
-    def model = body.model
-    def prompt_tokens = usage["prompt_tokens"]
-    def completion_tokens = usage["completion_tokens"]
-    def total_tokens = usage["total_tokens"]
-    def usage = body.usage || {}
-
-    private
-
-    def format_tool_calls(tools)
-      (tools || []).filter_map do |tool|
-        next unless tool.function
-        tool = {
-          id: tool.id,
-          name: tool.function.name,
-          arguments: JSON.parse(tool.function.arguments)
-        }
-        LLM::Object.new(tool)
-      end
-    end
-  end
-end