raif 1.4.0 → 1.5.0

data/app/models/raif/streaming_responses/bedrock.rb

@@ -3,6 +3,8 @@
 class Raif::StreamingResponses::Bedrock
 
   def initialize_new_message
+    @reasoning_content_blocks = {}
+
     # Initialize empty AWS response object
     @message = Aws::BedrockRuntime::Types::Message.new(
       role: "assistant",
@@ -62,9 +64,12 @@ class Raif::StreamingResponses::Bedrock
         )
 
         @message.content[index].tool_use.input += event.delta.tool_use.input
+      elsif event.delta.is_a?(Aws::BedrockRuntime::Types::ContentBlockDelta::ReasoningContent)
+        accumulate_reasoning_content(index, event.delta.reasoning_content)
       end
     when :content_block_stop
-      content_block = @message.content[event.content_block_index]
+      index = event.content_block_index
+      content_block = @message.content[index]
 
       if content_block&.tool_use&.input.is_a?(String)
         begin
@@ -73,6 +78,8 @@ class Raif::StreamingResponses::Bedrock
           # If parsing fails, leave as a string
         end
       end
+
+      finalize_reasoning_content(index)
     when :message_stop
       @response.stop_reason = event.stop_reason
     when :metadata
@@ -86,4 +93,56 @@ class Raif::StreamingResponses::Bedrock
     @response
   end
 
+private
+
+  def accumulate_reasoning_content(index, reasoning_delta)
+    reasoning_content = reasoning_content_for(index)
+    reasoning_content[:seen] = true
+
+    case reasoning_delta
+    when Aws::BedrockRuntime::Types::ReasoningContentBlockDelta::Text
+      reasoning_content[:text] << reasoning_delta.text.to_s
+    when Aws::BedrockRuntime::Types::ReasoningContentBlockDelta::Signature
+      reasoning_content[:signature] = reasoning_delta.signature
+    when Aws::BedrockRuntime::Types::ReasoningContentBlockDelta::RedactedContent
+      reasoning_content[:redacted_content] << reasoning_delta.redacted_content.to_s
+    else
+      reasoning_content[:unknown] = true
+    end
+  end
+
+  def finalize_reasoning_content(index)
+    reasoning_content = @reasoning_content_blocks.delete(index)
+    return unless reasoning_content&.dig(:seen)
+
+    @message.content[index] = Aws::BedrockRuntime::Types::ContentBlock::ReasoningContent.new(
+      reasoning_content: build_reasoning_content(reasoning_content)
+    )
+  end
+
+  def build_reasoning_content(reasoning_content)
+    if reasoning_content[:text].blank? && reasoning_content[:signature].blank? && reasoning_content[:redacted_content].present?
+      return Aws::BedrockRuntime::Types::ReasoningContentBlock::RedactedContent.new(
+        redacted_content: reasoning_content[:redacted_content]
+      )
+    end
+
+    Aws::BedrockRuntime::Types::ReasoningContentBlock::ReasoningText.new(
+      reasoning_text: Aws::BedrockRuntime::Types::ReasoningTextBlock.new(
+        text: reasoning_content[:text],
+        signature: reasoning_content[:signature]
+      )
+    )
+  end
+
+  def reasoning_content_for(index)
+    @reasoning_content_blocks[index] ||= {
+      seen: false,
+      text: +"",
+      signature: nil,
+      redacted_content: +"",
+      unknown: false
+    }
+  end
+
 end

data/app/models/raif/task.rb

@@ -11,6 +11,7 @@
 #  failed_at              :datetime
 #  llm_model_key          :string           not null
 #  prompt                 :text
+#  prompt_studio_run      :boolean          default(FALSE), not null
 #  raw_response           :text
 #  requested_language_key :string
 #  response_format        :integer          default("text"), not null
@@ -39,12 +40,16 @@
 #
 module Raif
   class Task < Raif::ApplicationRecord
+    prepend Raif::Concerns::HasPromptTemplates
+
     include Raif::Concerns::HasLlm
     include Raif::Concerns::HasRequestedLanguage
     include Raif::Concerns::HasAvailableModelTools
+    include Raif::Concerns::HasRuntimeDuration
     include Raif::Concerns::InvokesModelTools
     include Raif::Concerns::LlmResponseParsing
     include Raif::Concerns::LlmTemperature
+    include Raif::Concerns::LlmPromptCaching
     include Raif::Concerns::JsonSchemaDefinition
     include Raif::Concerns::RunWith
 
@@ -95,7 +100,7 @@ module Raif
     # @param files [Array] Optional array of Raif::ModelFileInput objects to include with the prompt.
     # @param args [Hash] Additional arguments to pass to the instance of the task that is created.
     # @return [Raif::Task, nil] The task instance that was created and run.
-    def self.run(creator: nil, available_model_tools: [], llm_model_key: nil, images: [], files: [], **args)
+    def self.run(creator: nil, available_model_tools: [], llm_model_key: nil, images: [], files: [], **args, &block)
       task = new(
         creator: creator,
         llm_model_key: llm_model_key,
@@ -107,7 +112,7 @@ module Raif
       )
 
       task.save!
-      task.run
+      task.run(&block)
       task
     rescue StandardError => e
       task&.failed!
@@ -126,18 +131,28 @@ module Raif
       task
     end
 
-    def run(skip_prompt_population: false)
+    def run(skip_prompt_population: false, &block)
       update_columns(started_at: Time.current) if started_at.nil?
 
       populate_prompts unless skip_prompt_population
 
+      streaming_block = if block_given?
+        proc do |model_completion, delta, sse_event|
+          update_columns(raw_response: model_completion.raw_response, updated_at: Time.current)
+          block.call(model_completion, delta, sse_event)
+        end
+      end
+
       mc = llm.chat(
         messages: messages,
         source: self,
         system_prompt: system_prompt,
         response_format: response_format.to_sym,
         available_model_tools: available_model_tools,
-        temperature: self.class.temperature
+        temperature: self.class.temperature,
+        anthropic_prompt_caching_enabled: self.class.anthropic_prompt_caching_enabled,
+        bedrock_prompt_caching_enabled: self.class.bedrock_prompt_caching_enabled,
+        &streaming_block
       )
 
       self.raif_model_completion = mc.becomes(Raif::ModelCompletion)
@@ -149,9 +164,9 @@ module Raif
       self
     end
 
-    def re_run
+    def re_run(&block)
       update_columns(started_at: Time.current)
-      run(skip_prompt_population: true)
+      run(skip_prompt_population: true, &block)
     end
 
     def messages
@@ -191,6 +206,15 @@ module Raif
       schema_for_instance(:json_response)
     end
 
+    # Returns additional attributes to assign when creating tasks in Prompt Studio
+    # (reruns, batch runs, and judge tasks). Override in your ApplicationTask or
+    # task subclass to include app-specific attributes.
+    #
+    # @return [Hash] additional attributes to assign to the new task
+    def prompt_studio_task_attributes
+      {}
+    end
+
     def build_prompt
       raise NotImplementedError, "Raif::Task subclasses must implement #build_prompt"
     end

data/app/views/layouts/raif/admin.html.erb

@@ -5,14 +5,23 @@
   <meta name="viewport" content="width=device-width,initial-scale=1">
   <%= csrf_meta_tags %>
   <%= csp_meta_tag %>
+  <%= action_cable_meta_tag if respond_to?(:action_cable_meta_tag) %>
 
   <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-T3c6CoIi6uLrA9TneNEoa7RxnatzjcDSCmG1MXxSR1GAsXEV/Dwwykc2MPK8M2HN" crossorigin="anonymous">
   <script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/js/bootstrap.bundle.min.js" integrity="sha384-C6RzsynM9kWDrMNeT87bh95OGNyZPhcTNXj1NW7RuBCsyN/o0jlpcV8Qyq46cDfL" crossorigin="anonymous" nonce="<%= request.content_security_policy_nonce %>"></script>
 
+  <link href="https://cdn.jsdelivr.net/npm/tom-select@2.4.3/dist/css/tom-select.bootstrap5.min.css" rel="stylesheet">
+  <script src="https://cdn.jsdelivr.net/npm/tom-select@2.4.3/dist/js/tom-select.complete.min.js" nonce="<%= request.content_security_policy_nonce %>"></script>
+
+  <% if respond_to?(:javascript_importmap_tags) %>
+    <%= javascript_importmap_tags "raif_admin" %>
+  <% else %>
+    <%= javascript_include_tag "raif_admin_sprockets", nonce: true %>
+  <% end %>
   <%= stylesheet_link_tag "raif_admin" %>
 </head>
 
-<body class="raif-admin">
+<body class="raif-admin" data-controller="raif--copy-to-clipboard">
   <nav class="navbar navbar-expand-md navbar-dark bg-dark">
     <div class="container-fluid">
       <a class="navbar-brand fw-bold" href="<%= raif.admin_tasks_path %>">
@@ -96,6 +105,27 @@
                 <%= t("raif.admin.common.agents") %>
               </a>
             </li>
+            <li class="nav-item">
+              <a class="nav-link <%= current_page?(raif.admin_llms_path) ? "active" : "" %>" href="<%= raif.admin_llms_path %>">
+                <span class="d-inline-block me-2">
+                  <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-currency-dollar" viewBox="0 0 16 16">
+                    <path d="M4 10.781c.148 1.667 1.513 2.85 3.591 3.003V15h1.043v-1.216c2.27-.179 3.678-1.438 3.678-3.3 0-1.59-.947-2.51-2.956-3.028l-.722-.187V3.467c1.122.11 1.879.714 2.07 1.616h1.47c-.166-1.6-1.54-2.748-3.54-2.875V1H7.591v1.233c-1.939.23-3.27 1.472-3.27 3.156 0 1.454.966 2.483 2.661 2.917l.61.162v4.031c-1.149-.17-1.94-.8-2.131-1.718H4zm3.391-3.836c-1.043-.263-1.6-.825-1.6-1.616 0-.944.704-1.641 1.8-1.828v3.495l-.2-.05zm1.591 1.872c1.287.323 1.852.859 1.852 1.769 0 1.097-.826 1.828-2.2 1.939V8.73l.348.086z" />
+                  </svg>
+                </span>
+                <%= t("raif.admin.common.llms") %>
+              </a>
+            </li>
+            <li class="nav-item">
+              <a class="nav-link <%= params[:controller].start_with?("raif/admin/prompt_studio") ? "active" : "" %>" href="<%= raif.admin_prompt_studio_tasks_path %>">
+                <span class="d-inline-block me-2">
+                  <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-pencil-square" viewBox="0 0 16 16">
+                    <path d="M15.502 1.94a.5.5 0 0 1 0 .706L14.459 3.69l-2-2L13.502.646a.5.5 0 0 1 .707 0l1.293 1.293zm-1.75 2.456-2-2L4.939 9.21a.5.5 0 0 0-.121.196l-.805 2.414a.25.25 0 0 0 .316.316l2.414-.805a.5.5 0 0 0 .196-.12l6.813-6.814z" />
+                    <path fill-rule="evenodd" d="M1 13.5A1.5 1.5 0 0 0 2.5 15h11a1.5 1.5 0 0 0 1.5-1.5v-6a.5.5 0 0 0-1 0v6a.5.5 0 0 1-.5.5h-11a.5.5 0 0 1-.5-.5v-11a.5.5 0 0 1 .5-.5H9a.5.5 0 0 0 0-1H2.5A1.5 1.5 0 0 0 1 2.5v11z" />
+                  </svg>
+                </span>
+                <%= t("raif.admin.prompt_studio.common.prompt_studio") %>
+              </a>
+            </li>
             <li class="nav-item">
               <a class="nav-link <%= current_page?(raif.admin_config_path) ? "active" : "" %>" href="<%= raif.admin_config_path %>">
                 <span class="d-inline-block me-2">

data/app/views/raif/admin/agents/_agent.html.erb

@@ -14,6 +14,7 @@
       <span class="badge bg-secondary"><%= t("raif.admin.common.pending") %></span>
     <% end %>
   </td>
+  <td><small class="text-muted"><%= agent.runtime_duration %></small></td>
   <td><%= agent.iteration_count %> / <%= agent.max_iterations %></td>
   <td>
     <% if agent.total_cost && agent.total_cost > 0 %>

data/app/views/raif/admin/agents/index.html.erb

@@ -2,6 +2,53 @@
 
 <div class="row">
   <div class="col-12">
+    <%= form_tag raif.admin_agents_path, method: :get, class: "mb-4" do %>
+      <div class="row align-items-end">
+        <div class="col-md-3">
+          <div class="form-group">
+            <label for="agent_type"><%= t("raif.admin.common.type") %></label>
+            <%= select_tag :agent_type,
+                  options_for_select(
+                    [[t("raif.admin.common.all"), "all"]] + @agent_types.map { |type| [type, type] },
+                    @selected_type
+                  ),
+                  { class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
+        <div class="col-md-3">
+          <div class="form-group">
+            <label for="status"><%= t("raif.admin.common.status") %></label>
+            <%= select_tag :status,
+                  options_for_select(
+                    [
+                      [t("raif.admin.common.all"), :all],
+                      [t("raif.admin.common.completed"), :completed],
+                      [t("raif.admin.common.failed"), :failed],
+                      [t("raif.admin.common.running"), :running],
+                      [t("raif.admin.common.pending"), :pending]
+                    ],
+                    @selected_status
+                  ),
+                  { class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
+        <div class="col-md-3">
+          <div class="form-group">
+            <label for="llm_model_key"><%= t("raif.admin.common.model") %></label>
+            <%= select_tag :llm_model_key,
+                  options_for_select(
+                    [[t("raif.admin.common.all"), ""]] + @llm_model_keys.map { |key| [key, key] },
+                    @selected_llm_model_key
+                  ),
+                  { class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
+        <div class="col-md-2">
+          <%= submit_tag t("raif.admin.common.filter"), class: "btn btn-primary" %>
+        </div>
+      </div>
+    <% end %>
+
     <% if @agents.any? %>
       <div class="table-responsive">
         <table class="table table-striped table-hover">
@@ -12,6 +59,7 @@
               <th><%= t("raif.admin.common.created_at") %></th>
               <th><%= t("raif.admin.common.task") %></th>
               <th><%= t("raif.admin.common.status") %></th>
+              <th><%= t("raif.admin.common.duration") %></th>
               <th><%= t("raif.admin.common.iterations") %></th>
               <th><%= t("raif.admin.common.total_cost") %></th>
               <th><%= t("raif.admin.common.final_answer") %></th>

data/app/views/raif/admin/agents/show.html.erb

@@ -36,6 +36,10 @@
         <% end %>
       </div>
     </div>
+    <div class="row mb-3">
+      <div class="col-md-3"><strong><%= t("raif.admin.common.duration") %>:</strong></div>
+      <div class="col-md-9"><%= @agent.runtime_duration %></div>
+    </div>
     <div class="row mb-3">
       <div class="col-md-3"><strong><%= t("raif.admin.common.iterations") %>:</strong></div>
       <div class="col-md-9"><%= @agent.iteration_count %> / <%= @agent.max_iterations %></div>

data/app/views/raif/admin/llms/index.html.erb

@@ -0,0 +1,110 @@
+<h1 class="my-4"><%= t("raif.admin.llms.index.title") %></h1>
+
+<div class="row">
+  <div class="col-12">
+    <%= form_tag raif.admin_llms_path, method: :get, class: "mb-4" do %>
+      <div class="row align-items-end">
+        <div class="col-md-4">
+          <div class="form-group">
+            <label for="names"><%= t("raif.admin.llms.index.name") %></label>
+            <%= select_tag "names[]",
+                  options_for_select(
+                    @llm_names.map { |n| [n, n] },
+                    @selected_names
+                  ),
+                  { id: "names", multiple: true, class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
+        <div class="col-md-4">
+          <div class="form-group">
+            <label for="providers"><%= t("raif.admin.llms.index.provider") %></label>
+            <%= select_tag "providers[]",
+                  options_for_select(
+                    @provider_names.map { |p| [p, p] },
+                    @selected_providers
+                  ),
+                  { id: "providers", multiple: true, class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
+        <div class="col-md-2">
+          <%= submit_tag t("raif.admin.common.filter"), class: "btn btn-primary" %>
+        </div>
+      </div>
+    <% end %>
+
+    <div class="card" data-controller="raif--table-search raif--sortable-table">
+      <div class="card-header bg-light">
+        <input type="text"
+               class="form-control form-control-sm"
+               placeholder="<%= t("raif.admin.llms.index.search_placeholder") %>"
+               data-raif--table-search-target="input"
+               autocomplete="off"
+               data-action="input->raif--table-search#filter">
+      </div>
+      <div class="card-body">
+        <% if @llms.any? %>
+          <table class="table table-sm table-hover mb-0">
+            <thead>
+              <tr>
+                <th role="button" data-action="click->raif--sortable-table#sort" data-raif--sortable-table-target="header" data-col-index="0" data-sort-type="string" style="cursor: pointer;">
+                  <%= t("raif.admin.llms.index.name") %>
+                </th>
+                <th role="button" data-action="click->raif--sortable-table#sort" data-raif--sortable-table-target="header" data-col-index="1" data-sort-type="string" style="cursor: pointer;">
+                  <%= t("raif.admin.llms.index.provider") %>
+                </th>
+                <th role="button" data-action="click->raif--sortable-table#sort" data-raif--sortable-table-target="header" data-col-index="2" data-sort-type="string" style="cursor: pointer;">
+                  <%= t("raif.admin.llms.index.api_name") %>
+                </th>
+                <th role="button" class="text-end" data-action="click->raif--sortable-table#sort" data-raif--sortable-table-target="header" data-col-index="3" data-sort-type="number" style="cursor: pointer;">
+                  <%= t("raif.admin.llms.index.input_cost") %>
+                </th>
+                <th role="button" class="text-end" data-action="click->raif--sortable-table#sort" data-raif--sortable-table-target="header" data-col-index="4" data-sort-type="number" style="cursor: pointer;">
+                  <%= t("raif.admin.llms.index.output_cost") %>
+                </th>
+                <th class="text-center"><%= t("raif.admin.common.default") %></th>
+              </tr>
+            </thead>
+            <tbody data-raif--sortable-table-target="body">
+              <% @llms.each do |llm| %>
+                <% input_cost = llm.input_token_cost.present? && llm.input_token_cost > 0 ? llm.input_token_cost * 1_000_000 : 0 %>
+                <% output_cost = llm.output_token_cost.present? && llm.output_token_cost > 0 ? llm.output_token_cost * 1_000_000 : 0 %>
+                <tr data-raif--table-search-target="row"
+                    data-searchable="<%= "#{llm.name} #{llm.class.name.demodulize} #{llm.api_name}".downcase %>">
+                  <td><%= llm.name %></td>
+                  <td><code><%= llm.class.name.demodulize %></code></td>
+                  <td><code><%= llm.api_name %></code></td>
+                  <td class="text-end" data-sort-value="<%= input_cost %>">
+                    <% if input_cost > 0 %>
+                      $<%= format("%.2f", input_cost) %>
+                    <% else %>
+                      <span class="text-muted">-</span>
+                    <% end %>
+                  </td>
+                  <td class="text-end" data-sort-value="<%= output_cost %>">
+                    <% if output_cost > 0 %>
+                      $<%= format("%.2f", output_cost) %>
+                    <% else %>
+                      <span class="text-muted">-</span>
+                    <% end %>
+                  </td>
+                  <td class="text-center">
+                    <% if llm.key.to_s == Raif.config.default_llm_model_key.to_s %>
+                      <span class="badge bg-primary"><%= t("raif.admin.llms.index.default_badge") %></span>
+                    <% end %>
+                  </td>
+                </tr>
+              <% end %>
+            </tbody>
+          </table>
+          <p class="text-muted mt-3 mb-0">
+            <small><%= t("raif.admin.llms.index.cost_note") %></small>
+          </p>
+        <% else %>
+          <div class="alert alert-warning mb-0">
+            <%= t("raif.admin.llms.index.no_llms") %>
+          </div>
+        <% end %>
+      </div>
+    </div>
+  </div>
+</div>

data/app/views/raif/admin/model_completions/_model_completion.html.erb

@@ -13,14 +13,10 @@
       <span class="badge bg-secondary"><%= t("raif.admin.common.pending") %></span>
     <% end %>
   </td>
+  <td><small class="text-muted"><%= model_completion.runtime_duration %></small></td>
+  <td><%= model_completion.cache_read_input_tokens.to_i > 0 ? number_with_delimiter(model_completion.cache_read_input_tokens) : "-" %></td>
+  <td><%= model_completion.cache_creation_input_tokens.to_i > 0 ? number_with_delimiter(model_completion.cache_creation_input_tokens) : "-" %></td>
   <td><%= model_completion.total_tokens ? number_with_delimiter(model_completion.total_tokens) : "-" %></td>
   <td><%= model_completion.total_cost ? number_to_currency(model_completion.total_cost, precision: 6) : "-" %></td>
-  <td>
-    <% if model_completion.citations.present? %>
-      <span class="badge bg-info"><%= model_completion.citations.length %></span>
-    <% else %>
-      <span class="text-muted">-</span>
-    <% end %>
-  </td>
   <td><small class="text-muted"><%= truncate(model_completion.raw_response, length: 100) %></small></td>
 </tr>

data/app/views/raif/admin/model_completions/index.html.erb

@@ -21,6 +21,17 @@
                   { class: "form-select" } %>
           </div>
         </div>
+        <div class="col-md-4">
+          <div class="form-group">
+            <label for="llm_model_key"><%= t("raif.admin.common.model") %></label>
+            <%= select_tag :llm_model_key,
+                  options_for_select(
+                    [[t("raif.admin.common.all"), ""]] + @llm_model_keys.map { |key| [key, key] },
+                    @selected_llm_model_key
+                  ),
+                  { class: "form-select", data: { controller: "raif--tom-select" } } %>
+          </div>
+        </div>
         <div class="col-md-2">
           <%= submit_tag t("raif.admin.common.filter"), class: "btn btn-primary" %>
         </div>
@@ -38,9 +49,11 @@
               <th><%= t("raif.admin.common.model") %></th>
               <th><%= t("raif.admin.common.response_format") %></th>
               <th><%= t("raif.admin.common.status") %></th>
+              <th><%= t("raif.admin.common.duration") %></th>
+              <th><%= t("raif.admin.common.cache_read_input_tokens") %></th>
+              <th><%= t("raif.admin.common.cache_creation_input_tokens") %></th>
               <th><%= t("raif.admin.common.total_tokens") %></th>
               <th><%= t("raif.admin.common.total_cost") %></th>
-              <th><%= t("raif.admin.common.citations") %></th>
               <th><%= t("raif.admin.common.response") %></th>
             </tr>
           </thead>