llmclt 1.0.1 → 1.0.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 664d39ff19d67ce78c661511aee009690e3f14c0654d6e1cca5c556a89cce060
-  data.tar.gz: 5536913892e091ad1b8435ee2a8ec508e96a9103ed707bd5eed05df742b61163
+  metadata.gz: 624b382b4d3337d9225a4c4d6d6892372f5a60444528dd2fc6a2f151eea81fae
+  data.tar.gz: 8a76cd4a7be9ec84f145e099cd3c0418428f23b4d58364422ff9319dc31d70ab
 SHA512:
-  metadata.gz: 9af77e20a63480c33f585359acf21179371726c497e01ff47248da5877f7b9f68db78deccb4806a290333d17276920c2fdf22c4d817de8a249159b501e770509
-  data.tar.gz: ad2524e8d9735e0a7b1249c0c93c380b6b6151eb9f22c99be9541256a937962a60cd17b7f70e75afc1c822eeb818a18aa1ee8333323d1494e3c53f1156860c3f
+  metadata.gz: 276766c1a32fce5c867b82575e49ef8fb361c8acf9ca06063c7adacc9aac877f1cbc5e75e351dc09da52754daea576972bedb75be5ec0a237e8b88e68e7c4b79
+  data.tar.gz: 7f2cc43c7cb87bd1f032031c562c47c8fed6be5b899a5e2eea6c72ac8804fce1330e13f871803e7565a0abeadb12734d3f98acbc90db5d32f339d4658845a03a

data/.rubocop.yml

@@ -1,12 +1,14 @@
 require:
   - rubocop-rake
+
+plugins:
   - rubocop-rspec
   - rubocop-performance
 
 inherit_from: .rubocop_todo.yml
 
 AllCops:
-  TargetRubyVersion: 2.7
+  TargetRubyVersion: 3.1
   NewCops: enable
   Exclude:
     - 'log/**/*'

data/CHANGELOG.md

@@ -1,5 +1,14 @@
 # CHANGELOG
 
+## 1.0.2
+
+- Rename a class from Request to Fetcher
+- Separate request and response classes for each mode
+- Add request jsonl output feature
+- Add batch request feature
+- Fix endpoint urls when location_id is global
+- Change required ruby version from 2.7 to 3.1
+
 ## 1.0.1
 
 - Add HTTP response code methods in Llmclt::Response

data/README.md

@@ -82,6 +82,47 @@ response = client.request(prompt, histories: histories)
 response.text
 ```
 
+### Output JSON Lines format
+
+```ruby
+client = Llmclt::Client.new()
+prompt = [
+  'Hello. This is Tom!. How are you?',
+  'Do you know my name?'
+]
+client.chat_jsonl(prompts)
+```
+then, it outputs JSON Lines like as follows:
+```json
+{"request":{"contents": [{"role": "user", "parts": [{"text": "Hello. This is Tom!. How are you?"}]}]}}
+{"request":{"contents": [{"role": "user", "parts": [{"text": "Do you know my name?"}]}]}}
+```
+
+### Batch mode
+
+```ruby
+client = Llmclt::Client.new(
+  project_id: project_id,
+  location_id: location_id,
+  model: model,
+  service_account_json: service_account_json
+)
+batch_job_name = 'BATCH_JOB_NAME'
+gcs_input_uri = 'gs://BUCKET_NAME/file.jsonl'
+gcs_output_uri = 'gs://BUCKET_NAME/'
+response = client.batch_request(batch_job_name, gcs_input_uri, gcs_output_uri)
+response.state
+=> 'JOB_STATE_PENDING'
+response.batch_job_id
+=> 'BATCH_JOB_ID'
+```
+then, you can poll for the status of the batch job using the `BATCH_JOB_ID` like as follows:
+```ruby
+response = client.batch_state_request('BATCH_JOB_ID')
+response.state
+=> 'JOB_STATE_SUCCEEDED'
+```
+
 ### Configuration
 
 ```ruby

data/lib/llmclt/client.rb

@@ -4,15 +4,27 @@ module Llmclt
   class Client
     def initialize(**kwargs)
       @config = Llmclt::Config.new(**kwargs)
-      @request = Llmclt::Request.new(@config)
+      @fetcher = Llmclt::Fetcher.new(@config)
     end
 
     def request(prompt, stream: false, histories: [])
-      @request.run(prompt, stream: stream, histories: histories)
+      @fetcher.run(prompt, stream: stream, histories: histories)
+    end
+
+    def chat_jsonl(prompts, histories: [])
+      Llmclt::Syntax::Chat.new(prompts, @config, histories: histories).to_jsonl
+    end
+
+    def batch_request(batch_job_name, gcs_input_uri, gcs_output_uri)
+      @fetcher.batch_run(batch_job_name, gcs_input_uri, gcs_output_uri)
+    end
+
+    def batch_state_request(batch_job_id)
+      @fetcher.batch_state_run(batch_job_id)
     end
 
     def shutdown
-      @request.shutdown
+      @fetcher.shutdown
     end
   end
 end

data/lib/llmclt/fetcher.rb

@@ -0,0 +1,113 @@
+# frozen_string_literal: true
+
+module Llmclt
+  class Fetcher
+    def initialize(config)
+      @config = config
+      @pool = Llmclt::Http::ConnectionPool.new
+    end
+
+    def run(prompt, stream: false, histories: [])
+      request = build_request(
+        request_mode(stream),
+        request_headers,
+        @config,
+        prompt,
+        histories
+      )
+      request_http(request)
+    end
+
+    def batch_run(batch_job_name, gcs_input_uri, gcs_output_uri)
+      request = Llmclt::Request::Batch.new(
+        request_headers,
+        @config,
+        batch_job_name: batch_job_name,
+        gcs_input_uri: gcs_input_uri,
+        gcs_output_uri: gcs_output_uri
+      )
+      request_http(request)
+    end
+
+    def batch_state_run(batch_job_id)
+      request = Llmclt::Request::Batch.new(
+        request_headers,
+        @config,
+        batch_job_id: batch_job_id
+      )
+      request_http(request)
+    end
+
+    def shutdown
+      @pool.shutdown
+    end
+
+    private
+
+    def request_http(request)
+      http = checkout_http(request.endpoint_uri)
+      response = http.start { |h| h.request(request.content) }
+      request.build_response(response)
+    end
+
+    def request_mode(stream)
+      if stream
+        'streaming'
+      else
+        'non_streaming'
+      end
+    end
+
+    def request_headers
+      authorizer.apply({ 'Content-Type': 'application/json' })
+    end
+
+    def authorizer
+      if @authorizer.nil? || @authorizer.expires_within?(60)
+        @authorizer = Google::Auth::ServiceAccountCredentials.make_creds(
+          json_key_io: StringIO.new(@config.service_account_json),
+          scope: 'https://www.googleapis.com/auth/cloud-platform'
+        )
+        @authorizer.fetch_access_token!
+      end
+      @authorizer
+    end
+
+    def build_request(request_mode, request_headers, config, prompt, histories)
+      case request_mode
+      when 'streaming'
+        Llmclt::Request::Streaming.new(
+          request_headers, config,
+          prompt: prompt,
+          histories: histories
+        )
+      when 'non_streaming'
+        Llmclt::Request::NonStreaming.new(
+          request_headers, config,
+          prompt: prompt,
+          histories: histories
+        )
+      else
+        raise NameError, "#{request_mode} is an invalid request mode"
+      end
+    end
+
+    def checkout_http(uri)
+      @pool.checkout(pool_name(uri)) { build_http(uri) }
+    end
+
+    def pool_name(uri)
+      "#{uri.scheme}_#{uri.host}_#{uri.port || uri.default_port}"
+    end
+
+    def build_http(uri)
+      http = Net::HTTP.new(uri.host, uri.port)
+      http.open_timeout = @config.open_timeout
+      http.read_timeout = @config.read_timeout
+      http.keep_alive_timeout = @config.keep_alive_timeout
+      http.use_ssl = true
+      http.verify_mode = OpenSSL::SSL::VERIFY_NONE
+      http
+    end
+  end
+end

data/lib/llmclt/request/base.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Request
+    class Base
+      def initialize(headers, config, **kwargs)
+        @headers = headers
+        @config = config
+        kwargs.each { |k, v| instance_variable_set(:"@#{k}", v) }
+      end
+
+      def endpoint_uri
+        raise NotImplementedError, "#{class_name}#{__method__} is not implemented"
+      end
+
+      def content
+        request = Net::HTTP::Post.new(endpoint_uri.request_uri)
+        @headers.each do |key, value|
+          request[key] = value
+        end
+        request.body = build_request_json
+        request
+      end
+
+      def build_response(response)
+        raise NotImplementedError, "#{class_name}#{__method__} is not implemented"
+      end
+
+      private
+
+      def endpoint_host
+        if @config.location_id == 'global'
+          'aiplatform.googleapis.com'
+        else
+          "#{@config.location_id}-aiplatform.googleapis.com"
+        end
+      end
+
+      def build_request_json
+        raise NotImplementedError, "#{class_name}#{__method__} is not implemented"
+      end
+    end
+  end
+end

data/lib/llmclt/request/batch.rb

@@ -0,0 +1,51 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Request
+    class Batch < Base
+      def endpoint_uri
+        url = "https://#{endpoint_host}/v1/projects/#{@config.project_id}/locations/#{@config.location_id}/batchPredictionJobs"
+        url = "#{url}/#{@batch_job_id}" if @batch_job_id
+        URI.parse(url)
+      end
+
+      def content
+        if @batch_job_id
+          request = Net::HTTP::Get.new(endpoint_uri.request_uri)
+          @headers.each do |key, value|
+            request[key] = value
+          end
+          request
+        else
+          super
+        end
+      end
+
+      def build_response(response)
+        Llmclt::Response::Batch.new(response)
+      end
+
+      private
+
+      def build_request_json
+        {
+          name: @batch_job_name,
+          displayName: @batch_job_name,
+          model: "publishers/google/models/#{@config.model}",
+          inputConfig: {
+            instancesFormat: 'jsonl',
+            gcsSource: {
+              uris: @gcs_input_uri
+            }
+          },
+          outputConfig: {
+            predictionsFormat: 'jsonl',
+            gcsDestination: {
+              outputUriPrefix: @gcs_output_uri
+            }
+          }
+        }.to_json
+      end
+    end
+  end
+end

data/lib/llmclt/request/chat.rb

@@ -0,0 +1,13 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Request
+    module Chat
+      private
+
+      def build_request_json
+        Llmclt::Syntax::Chat.new(@prompt, @config, histories: @histories).to_json
+      end
+    end
+  end
+end

data/lib/llmclt/request/non_streaming.rb

@@ -0,0 +1,19 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Request
+    class NonStreaming < Base
+      include Llmclt::Request::Chat
+
+      def endpoint_uri
+        URI.parse(
+          "https://#{endpoint_host}/v1/projects/#{@config.project_id}/locations/#{@config.location_id}/publishers/google/models/#{@config.model}:generateContent"
+        )
+      end
+
+      def build_response(response)
+        Llmclt::Response::NonStreaming.new(response)
+      end
+    end
+  end
+end

data/lib/llmclt/request/streaming.rb

@@ -0,0 +1,19 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Request
+    class Streaming < Base
+      include Llmclt::Request::Chat
+
+      def endpoint_uri
+        URI.parse(
+          "https://#{endpoint_host}/v1/projects/#{@config.project_id}/locations/#{@config.location_id}/publishers/google/models/#{@config.model}:streamGenerateContent"
+        )
+      end
+
+      def build_response(response)
+        Llmclt::Response::Streaming.new(response)
+      end
+    end
+  end
+end

data/lib/llmclt/response/base.rb

@@ -0,0 +1,28 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Response
+    class Base
+      attr_reader :response_orig, :response
+
+      def initialize(response)
+        @response_orig = response
+        @response = JSON.parse(response.body) if response.body
+      rescue JSON::ParserError
+        @response = nil
+      end
+
+      def status
+        response_orig.code.to_i
+      end
+
+      def success?
+        status == 200
+      end
+
+      def error?
+        status.between?(400, 599)
+      end
+    end
+  end
+end

data/lib/llmclt/response/batch.rb

@@ -0,0 +1,19 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Response
+    class Batch < Base
+      def batch_job_id
+        @response['name'].split('/').last
+      end
+
+      def state
+        @response['state']
+      end
+
+      def state_succeeded?
+        @response['state'] == 'JOB_STATE_SUCCEEDED'
+      end
+    end
+  end
+end

data/lib/llmclt/response/chat.rb

@@ -0,0 +1,15 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Response
+    module Chat
+      def text
+        texts.join
+      end
+
+      def texts
+        raise NotImplementedError, "#{class_name}#{__method__} is not implemented"
+      end
+    end
+  end
+end

data/lib/llmclt/response/non_streaming.rb

@@ -0,0 +1,15 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Response
+    class NonStreaming < Base
+      include Llmclt::Response::Chat
+
+      def texts
+        candidates = response['candidates']
+        parts = candidates.map { |candidate| candidate['content']['parts'] }.flatten
+        parts.map { |part| part['text'] }
+      end
+    end
+  end
+end

data/lib/llmclt/response/streaming.rb

@@ -0,0 +1,15 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Response
+    class Streaming < Base
+      include Llmclt::Response::Chat
+
+      def texts
+        candidates = response.map { |res| res['candidates'] }.flatten
+        parts = candidates.map { |candidate| candidate['content']['parts'] }.flatten
+        parts.map { |part| part['text'] }
+      end
+    end
+  end
+end

data/lib/llmclt/syntax/chat.rb

@@ -0,0 +1,79 @@
+# frozen_string_literal: true
+
+module Llmclt
+  module Syntax
+    class Chat
+      attr_reader :prompts, :config, :histories
+      attr_reader :params
+
+      def initialize(prompts, config, histories: [])
+        @prompts = Array(prompts)
+        @config = config
+        @histories = histories
+        @params = build_params
+      end
+
+      def to_json(*_args)
+        params.size == 1 ? params.first.to_json : params.to_json
+      end
+
+      def to_jsonl
+        params.map { |json| { request: json } }.map(&:to_json).join("\n")
+      end
+
+      private
+
+      def build_params
+        prompts.each_with_object([]) do |prompt, prms|
+          prm = {}
+          prm.merge!(build_contents(prompt))
+          prm.merge!(build_system_instruction)
+          prms << prm
+        end
+      end
+
+      def build_contents(prompt)
+        content = { contents: [] }
+        content.merge!(config.safety_config.build_request_content)
+        content.merge!(config.generation_config.build_request_content)
+        content[:contents].concat(build_history_contents(histories))
+        content[:contents] << build_user_prompt(prompt)
+        content
+      end
+
+      def build_history_contents(histories)
+        histories.map do |history|
+          {
+            role: history[:role],
+            parts: [{ text: history[:text] }]
+          }
+        end
+      end
+
+      def build_user_prompt(prompt)
+        {
+          role: 'user',
+          parts: [
+            {
+              text: prompt
+            }
+          ]
+        }
+      end
+
+      def build_system_instruction
+        return {} if config.system_instruction_prompt.nil? || config.system_instruction_prompt.empty?
+
+        {
+          systemInstruction: {
+            parts: [
+              {
+                text: config.system_instruction_prompt
+              }
+            ]
+          }
+        }
+      end
+    end
+  end
+end

data/lib/llmclt/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Llmclt
-  VERSION = '1.0.1'
+  VERSION = '1.0.2'
 end

data/lib/llmclt.rb

@@ -7,10 +7,20 @@ require_relative 'llmclt/config'
 require_relative 'llmclt/config/safety_config'
 require_relative 'llmclt/config/generation_config'
 require_relative 'llmclt/client'
-require_relative 'llmclt/request'
+require_relative 'llmclt/fetcher'
+require_relative 'llmclt/syntax/chat'
+require_relative 'llmclt/request/base'
+require_relative 'llmclt/request/chat'
+require_relative 'llmclt/request/non_streaming'
+require_relative 'llmclt/request/streaming'
+require_relative 'llmclt/request/batch'
 require_relative 'llmclt/http/connection'
 require_relative 'llmclt/http/connection_pool'
-require_relative 'llmclt/response'
+require_relative 'llmclt/response/base'
+require_relative 'llmclt/response/chat'
+require_relative 'llmclt/response/non_streaming'
+require_relative 'llmclt/response/streaming'
+require_relative 'llmclt/response/batch'
 require_relative 'llmclt/version'
 
 module Llmclt

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llmclt
 version: !ruby/object:Gem::Version
-  version: 1.0.1
+  version: 1.0.2
 platform: ruby
 authors:
 - MasatoMiyoshi
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2025-01-28 00:00:00.000000000 Z
+date: 2025-05-16 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: activesupport
@@ -58,10 +58,20 @@ files:
 - lib/llmclt/config.rb
 - lib/llmclt/config/generation_config.rb
 - lib/llmclt/config/safety_config.rb
+- lib/llmclt/fetcher.rb
 - lib/llmclt/http/connection.rb
 - lib/llmclt/http/connection_pool.rb
-- lib/llmclt/request.rb
-- lib/llmclt/response.rb
+- lib/llmclt/request/base.rb
+- lib/llmclt/request/batch.rb
+- lib/llmclt/request/chat.rb
+- lib/llmclt/request/non_streaming.rb
+- lib/llmclt/request/streaming.rb
+- lib/llmclt/response/base.rb
+- lib/llmclt/response/batch.rb
+- lib/llmclt/response/chat.rb
+- lib/llmclt/response/non_streaming.rb
+- lib/llmclt/response/streaming.rb
+- lib/llmclt/syntax/chat.rb
 - lib/llmclt/version.rb
 - sig/llmclt.rbs
 homepage: https://github.com/MasatoMiyoshi/llmclt
@@ -79,7 +89,7 @@ required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
-      version: 2.7.0
+      version: 3.1.0
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="

data/lib/llmclt/request.rb

@@ -1,128 +0,0 @@
-# frozen_string_literal: true
-
-module Llmclt
-  class Request
-    def initialize(config)
-      @config = config
-      @pool = Llmclt::Http::ConnectionPool.new
-    end
-
-    def run(prompt, stream: false, histories: [])
-      uri = request_uri(stream)
-      request = build_request(uri, prompt, histories)
-      http = checkout_http(uri)
-      response = http.start { |h| h.request(request) }
-      Llmclt::Response.new(response)
-    end
-
-    def shutdown
-      @pool.shutdown
-    end
-
-    private
-
-    def request_uri(stream)
-      if stream
-        URI.parse("#{endpoint_url}:streamGenerateContent")
-      else
-        URI.parse("#{endpoint_url}:generateContent")
-      end
-    end
-
-    def endpoint_url
-      "https://#{@config.location_id}-aiplatform.googleapis.com/v1/projects/#{@config.project_id}/locations/#{@config.location_id}/publishers/google/models/#{@config.model}"
-    end
-
-    def request_headers
-      authorizer.apply({ 'Content-Type': 'application/json' })
-    end
-
-    def authorizer
-      if @authorizer.nil? || @authorizer.expires_within?(60)
-        @authorizer = Google::Auth::ServiceAccountCredentials.make_creds(
-          json_key_io: StringIO.new(@config.service_account_json),
-          scope: 'https://www.googleapis.com/auth/cloud-platform'
-        )
-        @authorizer.fetch_access_token!
-      end
-      @authorizer
-    end
-
-    def checkout_http(uri)
-      @pool.checkout(pool_name(uri)) { build_http(uri) }
-    end
-
-    def pool_name(uri)
-      "#{uri.scheme}_#{uri.host}_#{uri.port || uri.default_port}"
-    end
-
-    def build_http(uri)
-      http = Net::HTTP.new(uri.host, uri.port)
-      http.open_timeout = @config.open_timeout
-      http.read_timeout = @config.read_timeout
-      http.keep_alive_timeout = @config.keep_alive_timeout
-      http.use_ssl = true
-      http.verify_mode = OpenSSL::SSL::VERIFY_NONE
-      http
-    end
-
-    def build_request(uri, prompt, histories)
-      request = Net::HTTP::Post.new(uri.request_uri)
-      request_headers.each do |key, value|
-        request[key] = value
-      end
-      request.body = build_request_json(prompt, histories)
-      request
-    end
-
-    def build_request_json(prompt, histories)
-      req = {}
-      req.merge!(build_request_contents(prompt, histories))
-      req.merge!(build_request_system_instruction)
-      req.to_json
-    end
-
-    def build_request_contents(prompt, histories)
-      content = { contents: [] }
-      content.merge!(@config.safety_config.build_request_content)
-      content.merge!(@config.generation_config.build_request_content)
-      content[:contents].concat(build_request_history_contents(histories))
-      content[:contents] << build_request_user_prompt(prompt)
-      content
-    end
-
-    def build_request_history_contents(histories)
-      histories.map do |history|
-        {
-          role: history[:role],
-          parts: [{ text: history[:text] }]
-        }
-      end
-    end
-
-    def build_request_user_prompt(prompt)
-      {
-        role: 'user',
-        parts: [
-          {
-            text: prompt
-          }
-        ]
-      }
-    end
-
-    def build_request_system_instruction
-      return {} if @config.system_instruction_prompt.nil? || @config.system_instruction_prompt.empty?
-
-      {
-        systemInstruction: {
-          parts: [
-            {
-              text: @config.system_instruction_prompt
-            }
-          ]
-        }
-      }
-    end
-  end
-end

data/lib/llmclt/response.rb

@@ -1,46 +0,0 @@
-# frozen_string_literal: true
-
-module Llmclt
-  class Response
-    attr_reader :response_orig, :response
-
-    def initialize(response)
-      @response_orig = response
-      @response = JSON.parse(response.body) if response.body
-    rescue JSON::ParserError
-      @response = nil
-    end
-
-    def status
-      response_orig.code.to_i
-    end
-
-    def success?
-      status == 200
-    end
-
-    def error?
-      status >= 400 && status <= 599
-    end
-
-    def text
-      texts.join
-    end
-
-    def texts
-      candidates = responses.map { |res| res['candidates'] }.flatten
-      parts = candidates.map { |candidate| candidate['content']['parts'] }.flatten
-      parts.map { |part| part['text'] }
-    end
-
-    def stream?
-      response.is_a?(Array)
-    end
-
-    private
-
-    def responses
-      stream? ? response : [response]
-    end
-  end
-end