completion-kit 0.26.5 → 0.26.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 870c7454b6181023648742b79a1682b09137cdd8480074b8c10c9382c1e0eb4c
-  data.tar.gz: b4faee283f8da861406d36c90ef201ea7bf35d4d725ec3103b42f9025d6b3beb
+  metadata.gz: a1d7db0abdd990413e60ba5ef164580ef34485396f02598336ddca4e2ce39f4d
+  data.tar.gz: 7a8563ba729f30c7296fefd6aab9b3de6b6aee8c1b7823739ed96d545424f96c
 SHA512:
-  metadata.gz: 0f20cad2387e3c95fa2a00b6caebde697de4c10d21eb7b44f71b2a47bfcfa60100dc5a94041aa435a5de470cff2a7b9a214e976c4c6a4f2a6d83f0f4baf71b4f
-  data.tar.gz: a713d5273431e17b145a14f85645aa437c143f2d86e93f6401ddd0bd7d6ade437004870aa5c03da39967003ec5d4429d5977df19b1ea981bf70db3c042475a5b
+  metadata.gz: e3b52c0043a89ae3fdf78735291fa4caffdd043c98f5ad35c99ea2d9969ba8c3b284ae50d24cde31f528e35cf4a51172b946d150116bc525e71fe02658528108
+  data.tar.gz: 946d48be8c7d204cc94d47f97cc0c71856199226e8602d3e94df293204d4cfe87dff50a70d2b71f672e0df0cf82dc7d061de4908ba2dd3599e87cc5640eb59e1

data/app/services/completion_kit/azure_foundry_client.rb

@@ -12,12 +12,21 @@ module CompletionKit
       model = options[:model]
       max_tokens = options[:max_tokens] || 1000
       temperature = options[:temperature] || 0.7
-
-      response = post_chat(model: model, prompt: prompt, max_tokens: max_tokens, temperature: temperature)
-
-      if response.status == 400 && temperature_unsupported?(response.body)
-        @temperature_dropped = true
-        response = post_chat(model: model, prompt: prompt, max_tokens: max_tokens, temperature: nil)
+      max_completion = false
+
+      response = post_chat(model: model, prompt: prompt, max_tokens: max_tokens, temperature: temperature, max_completion: max_completion)
+
+      2.times do
+        break unless response.status == 400
+        if !max_completion && max_tokens_unsupported?(response.body)
+          max_completion = true
+        elsif !temperature.nil? && temperature_unsupported?(response.body)
+          @temperature_dropped = true
+          temperature = nil
+        else
+          break
+        end
+        response = post_chat(model: model, prompt: prompt, max_tokens: max_tokens, temperature: temperature, max_completion: max_completion)
       end
 
       if response.status == 429
@@ -49,12 +58,12 @@ module CompletionKit
       return [] unless configured?
       return [] unless ProviderEndpoint.safe?(api_endpoint)
 
-      response = build_connection(azure_base_url).get(models_path) do |req|
+      response = build_connection(azure_base_url).get(models_url) do |req|
         req.headers["api-key"] = api_key
       end
       return [] unless response.success?
 
-      JSON.parse(response.body).fetch("data", []).map { |entry| { id: entry["id"], name: entry["id"] } }
+      parse_models(response.body)
     rescue StandardError
       []
     end
@@ -92,13 +101,33 @@ module CompletionKit
       api_version.blank?
     end
 
-    def models_path
-      v1_mode? ? "/openai/v1/models" : "/openai/deployments?api-version=#{api_version}"
+    def foundry_project?
+      api_endpoint.to_s.include?("/api/projects/")
     end
 
-    def post_chat(model:, prompt:, max_tokens:, temperature:)
-      body = { messages: [{ role: "user", content: prompt }], max_tokens: max_tokens }
+    def models_url
+      if foundry_project?
+        "#{azure_base_url}/deployments?api-version=v1"
+      elsif v1_mode?
+        "/openai/v1/models"
+      else
+        "/openai/deployments?api-version=#{api_version}"
+      end
+    end
+
+    def parse_models(body)
+      json = JSON.parse(body)
+      if foundry_project?
+        json.fetch("value", []).map { |d| { id: d["name"], name: d["name"] } }
+      else
+        json.fetch("data", []).map { |e| { id: e["id"], name: e["id"] } }
+      end
+    end
+
+    def post_chat(model:, prompt:, max_tokens:, temperature:, max_completion: false)
+      body = { messages: [{ role: "user", content: prompt }] }
       body[:model] = model if v1_mode?
+      body[max_completion ? :max_completion_tokens : :max_tokens] = max_tokens
       body[:temperature] = temperature unless temperature.nil?
 
       build_connection(azure_base_url, timeout: 30, open_timeout: 5).post do |req|
@@ -113,5 +142,10 @@ module CompletionKit
       s = body.to_s
       s.include?("temperature") && (s.include?("deprecated") || s.include?("not supported") || s.include?("Unsupported parameter"))
     end
+
+    def max_tokens_unsupported?(body)
+      s = body.to_s
+      s.include?("max_tokens") && (s.include?("max_completion_tokens") || s.include?("not supported") || s.include?("Unsupported parameter"))
+    end
   end
 end

data/app/services/completion_kit/model_discovery_service.rb

@@ -156,17 +156,26 @@ module CompletionKit
     def fetch_azure_foundry_models
       raise DiscoveryError, "An Azure endpoint URL is required." if @api_endpoint.blank?
 
-      response = fetch_connection(azure_base_url).get(azure_models_path) do |req|
+      response = fetch_connection(azure_base_url).get(azure_foundry_project? ? "#{azure_base_url}/deployments?api-version=v1" : azure_models_path) do |req|
         req.headers["api-key"] = @api_key
       end
       raise DiscoveryError, azure_error_message(response) unless response.success?
-      JSON.parse(response.body).fetch("data", []).map { |e| { id: e["id"], display_name: e["id"] } }
+      body = JSON.parse(response.body)
+      if azure_foundry_project?
+        body.fetch("value", []).map { |d| { id: d["name"], display_name: d["name"] } }
+      else
+        body.fetch("data", []).map { |e| { id: e["id"], display_name: e["id"] } }
+      end
     end
 
     def azure_v1_mode?
       @api_version.blank?
     end
 
+    def azure_foundry_project?
+      @api_endpoint.to_s.include?("/api/projects/")
+    end
+
     def azure_models_path
       azure_v1_mode? ? "/openai/v1/models" : "/openai/deployments?api-version=#{@api_version}"
     end
@@ -177,8 +186,16 @@ module CompletionKit
 
     def azure_error_message(response)
       detail = extract_provider_error_message(response.body)
-      hint = azure_v1_mode? ? "Check the endpoint base URL." : "Check the endpoint base URL and api-version."
-      path = azure_v1_mode? ? "/openai/v1/models" : "/openai/deployments"
+      if azure_foundry_project?
+        path = "/deployments"
+        hint = "Check the project endpoint URL."
+      elsif azure_v1_mode?
+        path = "/openai/v1/models"
+        hint = "Check the endpoint base URL."
+      else
+        path = "/openai/deployments"
+        hint = "Check the endpoint base URL and api-version."
+      end
       with_detail("Azure did not return a model list at #{path} (#{response.status}). #{hint}", detail)
     end
 
@@ -437,8 +454,17 @@ module CompletionKit
         f.request :retry, max: 1, interval: 0.5
         f.adapter Faraday.default_adapter
       end
-      body = { messages: [{ role: "user", content: input }], max_tokens: max_tokens }
+      response = azure_probe_post(conn, model_id, input, max_tokens, max_completion: false)
+      if response.status == 400 && azure_max_tokens_unsupported?(response.body)
+        response = azure_probe_post(conn, model_id, input, max_tokens, max_completion: true)
+      end
+      response
+    end
+
+    def azure_probe_post(conn, model_id, input, max_tokens, max_completion:)
+      body = { messages: [{ role: "user", content: input }] }
       body[:model] = model_id if azure_v1_mode?
+      body[max_completion ? :max_completion_tokens : :max_tokens] = max_tokens
       conn.post do |req|
         req.url(azure_v1_mode? ? "/openai/v1/chat/completions" : "/openai/deployments/#{model_id}/chat/completions?api-version=#{@api_version}")
         req.headers["Content-Type"] = "application/json"
@@ -446,5 +472,10 @@ module CompletionKit
         req.body = body.to_json
       end
     end
+
+    def azure_max_tokens_unsupported?(body)
+      s = body.to_s
+      s.include?("max_tokens") && (s.include?("max_completion_tokens") || s.include?("not supported") || s.include?("Unsupported parameter"))
+    end
   end
 end

data/app/views/completion_kit/provider_credentials/edit.html.erb

@@ -1,5 +1,5 @@
 <ol class="ck-breadcrumb">
-  <li><%= link_to "Settings", provider_credentials_path %></li>
+  <li><%= link_to "Providers", provider_credentials_path %></li>
   <li>Edit</li>
 </ol>
 

data/app/views/completion_kit/provider_credentials/new.html.erb

@@ -1,5 +1,5 @@
 <ol class="ck-breadcrumb">
-  <li><%= link_to "Settings", provider_credentials_path %></li>
+  <li><%= link_to "Providers", provider_credentials_path %></li>
   <li>New</li>
 </ol>
 

data/app/views/completion_kit/runs/show.html.erb

@@ -64,6 +64,12 @@
       <span class="ck-run-config__none">None</span>
     <% end %>
   </div>
+  <% if @run.prompt&.llm_model.present? %>
+    <div class="ck-run-config__row">
+      <span class="ck-run-config__key">Model</span>
+      <span style="text-transform: none;"><%= @run.prompt.llm_model %></span>
+    </div>
+  <% end %>
   <% if @run.judge_model.present? %>
     <div class="ck-run-config__row">
       <span class="ck-run-config__key">Judge</span>

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.26.5"
+  VERSION = "0.26.6"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.26.5
+  version: 0.26.6
 platform: ruby
 authors:
 - Damien Bastin