Merge pull request trustyai-explainability#141 from trustyai-explainability/rhoaieng-55152

Rename 'stubs' to 'prompts' in report UI labels

Author: saichandrapandraju

src/llama_stack_provider_trustyai_garak/resources/art_report.jinja2

@@ -157,10 +157,10 @@
                                             <th role="columnheader" scope="col">Intent</th>
                                             <th role="columnheader" scope="col">Total Attempts</th>
                                             {% if probe.is_baseline %}
-                                                <th role="columnheader" scope="col">Unsafe Stubs</th>
+                                                <th role="columnheader" scope="col">Unsafe Prompts</th>
                                                 <th role="columnheader" scope="col">Unsafe Rate</th>
                                             {% else %}
-                                                <th role="columnheader" scope="col">Unsafe Stubs</th>
+                                                <th role="columnheader" scope="col">Unsafe Prompts</th>
                                                 <th role="columnheader" scope="col">Attack Success Rate</th>
                                             {% endif %}
                                         </tr>
@@ -171,13 +171,13 @@
                                                 <td role="cell" data-label="Intent">{{ row.intent_name }}</td>
                                                 <td role="cell" data-label="Total Attempts">{{ row.total_attacks }}</td>
                                                 {% if probe.is_baseline %}
-                                                    <td role="cell" data-label="Unsafe Stubs">{{ row.complied_attacks }}
+                                                    <td role="cell" data-label="Unsafe Prompts">{{ row.complied_attacks }}
                                                         / {{ row.total_attacks }}</td>
                                                     <td role="cell"
                                                         data-label="Unsafe Rate">{{ row.asr }}%
                                                     </td>
                                                 {% else %}
-                                                    <td role="cell" data-label="Unsafe Stubs">{{ row.jailbroken_stubs }}
+                                                    <td role="cell" data-label="Unsafe Prompts">{{ row.jailbroken_stubs }}
                                                         / {{ row.baseline_stubs }}</td>
                                                     <td role="cell"
                                                         data-label="Attack Success Rate">{{ row.asr }}%

src/llama_stack_provider_trustyai_garak/resources/vega_chart_behaviour_by_intent.json

@@ -91,7 +91,7 @@
       "aggregate": "max",
       "type": "quantitative",
       "legend": {
-        "title": "Baseline Stubs",
+        "title": "Baseline Prompts",
         "format": "d"
       }
     },
@@ -119,7 +119,7 @@
       },
       {
         "field": "stub",
-        "title": "Stub"
+        "title": "Prompt"
       },
       {
         "field": "probe_name",
@@ -132,7 +132,7 @@
       {
         "field": "baseline_stubs",
         "type": "quantitative",
-        "title": "Baseline Stubs"
+        "title": "Baseline Prompts"
       }
     ]
   },

src/llama_stack_provider_trustyai_garak/resources/vega_chart_behaviour_by_probe.json

@@ -184,15 +184,15 @@
           {
             "field": "baseline_stubs",
             "type": "quantitative",
-            "title": "Baseline Stubs"
+            "title": "Baseline Prompts"
           },
           {
             "field": "intent_name",
             "title": "Intent"
           },
           {
             "field": "stub",
-            "title": "Stub"
+            "title": "Prompt"
           },
           {
             "field": "outcome",

src/llama_stack_provider_trustyai_garak/result_utils.py

@@ -1069,8 +1069,8 @@ def high_level_stats(attacks_by_intent_data):
 
     return [
         {"label": "Total attempts", "value": total_attempts},
-        {"label": "Unsafe stubs", "value": unsafe_count},
-        {"label": "Safe stubs", "value": safe_count},
+        {"label": "Unsafe prompts", "value": unsafe_count},
+        {"label": "Safe prompts", "value": safe_count},
         {"label": "Attack success rate", "value": format(success_rate, ".0f") + "%"},
     ]
 

tests/test_utils.py

@@ -741,8 +741,8 @@ def test_high_level_stats(self):
 
         stats = {s["label"]: s["value"] for s in result}
         assert stats["Total attempts"] == 5  # 1+1+1+2
-        assert stats["Unsafe stubs"] == 2  # stub_a + stub_d jailbroken across all probes
-        assert stats["Safe stubs"] == 2  # stub_b and stub_c never jailbroken
+        assert stats["Unsafe prompts"] == 2  # stub_a + stub_d jailbroken across all probes
+        assert stats["Safe prompts"] == 2  # stub_b and stub_c never jailbroken
         assert stats["Attack success rate"] == "50%"  # 2/4 unique stubs jailbroken
 
     def test_vega_data_includes_intent_name(self):
@@ -894,8 +894,8 @@ def test_intents_aggregates_match_high_level_stats(self):
         intents_metrics = calculate_intents_aggregates(all_raw)
 
         assert intents_metrics["total_attempts"] == art_dict["Total attempts"]
-        assert intents_metrics["unsafe_stubs"] == art_dict["Unsafe stubs"]
-        assert intents_metrics["safe_stubs"] == art_dict["Safe stubs"]
+        assert intents_metrics["unsafe_stubs"] == art_dict["Unsafe prompts"]
+        assert intents_metrics["safe_stubs"] == art_dict["Safe prompts"]
         expected_rate = art_dict["Attack success rate"].replace("%", "")
         assert format(intents_metrics["attack_success_rate"], ".0f") == expected_rate