From a884fc9676d9b5456cdcab4dfbdf4743aed53a41 Mon Sep 17 00:00:00 2001 From: saichandrapandraju Date: Sun, 1 Mar 2026 21:40:25 -0700 Subject: [PATCH 1/2] update garak provider configmap with supported benchmarks --- config/configmaps/evalhub/provider-garak.yaml | 109 +++++++++++------- 1 file changed, 70 insertions(+), 39 deletions(-) diff --git a/config/configmaps/evalhub/provider-garak.yaml b/config/configmaps/evalhub/provider-garak.yaml index cdd97aa3f..7fcd6d048 100644 --- a/config/configmaps/evalhub/provider-garak.yaml +++ b/config/configmaps/evalhub/provider-garak.yaml @@ -16,7 +16,8 @@ data: image: $(evalhub-provider-garak-image) entrypoint: - python - - /opt/app-root/src/main.py + - -m + - llama_stack_provider_trustyai_garak.evalhub cpu_request: 100m memory_request: 128Mi cpu_limit: 500m @@ -26,55 +27,85 @@ data: value: VALUE local: null benchmarks: - - id: toxicity - name: Toxicity Detection - description: Tests model's tendency to generate toxic content + - id: trustyai_garak::owasp_llm_top10 + name: OWASP LLM Top 10 + description: OWASP Top 10 for Large Language Model Applications + category: security + metrics: + - attack_success_rate + tags: + - security + - owasp + - red_team + - id: trustyai_garak::avid + name: AVID Taxonomy + description: AI Vulnerability and Incident Database - All vulnerabilities + category: security + metrics: + - attack_success_rate + tags: + - security + - avid + - red_team + - id: trustyai_garak::avid_security + name: AVID Security Taxonomy + description: AI Vulnerability and Incident Database - Security vulnerabilities + category: security + metrics: + - attack_success_rate + tags: + - security + - avid + - red_team + - id: trustyai_garak::avid_ethics + name: AVID Ethics Taxonomy + description: AI Vulnerability and Incident Database - Ethical concerns category: safety metrics: - - toxicity_rate - - severity_score - num_few_shot: 0 - dataset_size: 500 + - attack_success_rate tags: - safety - - toxicity + - ethics + - avid - red_team - - id: bias_detection - name: Bias Detection - description: Evaluates model for various forms of bias - category: fairness + - id: trustyai_garak::avid_performance + name: AVID Performance Taxonomy + description: AI Vulnerability and Incident Database - Performance issues + category: performance metrics: - - bias_score - - demographic_parity - num_few_shot: 0 - dataset_size: 1000 + - attack_success_rate tags: - - fairness - - bias - - demographic - - id: pii_leakage - name: PII Leakage - description: Tests for personally identifiable information leakage - category: privacy + - performance + - avid + - red_team + - id: trustyai_garak::quality + name: Quality Issues + description: Common quality issues like Violence, Profanity, Toxicity, Hate Speech, Integrity, etc. + category: safety metrics: - - pii_leak_rate - - sensitivity_score - num_few_shot: 0 - dataset_size: 300 + - attack_success_rate + tags: + - safety + - quality + - toxicity + - red_team + - id: trustyai_garak::cwe + name: Common Weakness Enumeration + description: Common Weakness Enumeration - Software security weaknesses + category: security + metrics: + - attack_success_rate tags: - - privacy - - pii - security - - id: prompt_injection - name: Prompt Injection - description: Tests resilience against prompt injection attacks + - cwe + - red_team + - id: trustyai_garak::quick + name: Quick Scan + description: Quick scan with a single probe for testing and validation category: security metrics: - - injection_success_rate - - defense_effectiveness - num_few_shot: 0 - dataset_size: 200 + - attack_success_rate tags: - security - - injection - - adversarial + - quick + - red_team From 046a1472546a8ebeb1d7bb991aef5fd9c101feba Mon Sep 17 00:00:00 2001 From: saichandrapandraju Date: Mon, 2 Mar 2026 07:55:24 -0700 Subject: [PATCH 2/2] rename garak benchmarks from trustyai_garak::x -> x --- config/configmaps/evalhub/provider-garak.yaml | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/config/configmaps/evalhub/provider-garak.yaml b/config/configmaps/evalhub/provider-garak.yaml index 7fcd6d048..0bccdfc29 100644 --- a/config/configmaps/evalhub/provider-garak.yaml +++ b/config/configmaps/evalhub/provider-garak.yaml @@ -27,7 +27,7 @@ data: value: VALUE local: null benchmarks: - - id: trustyai_garak::owasp_llm_top10 + - id: owasp_llm_top10 name: OWASP LLM Top 10 description: OWASP Top 10 for Large Language Model Applications category: security @@ -37,7 +37,7 @@ data: - security - owasp - red_team - - id: trustyai_garak::avid + - id: avid name: AVID Taxonomy description: AI Vulnerability and Incident Database - All vulnerabilities category: security @@ -47,7 +47,7 @@ data: - security - avid - red_team - - id: trustyai_garak::avid_security + - id: avid_security name: AVID Security Taxonomy description: AI Vulnerability and Incident Database - Security vulnerabilities category: security @@ -57,7 +57,7 @@ data: - security - avid - red_team - - id: trustyai_garak::avid_ethics + - id: avid_ethics name: AVID Ethics Taxonomy description: AI Vulnerability and Incident Database - Ethical concerns category: safety @@ -68,7 +68,7 @@ data: - ethics - avid - red_team - - id: trustyai_garak::avid_performance + - id: avid_performance name: AVID Performance Taxonomy description: AI Vulnerability and Incident Database - Performance issues category: performance @@ -78,7 +78,7 @@ data: - performance - avid - red_team - - id: trustyai_garak::quality + - id: quality name: Quality Issues description: Common quality issues like Violence, Profanity, Toxicity, Hate Speech, Integrity, etc. category: safety @@ -89,7 +89,7 @@ data: - quality - toxicity - red_team - - id: trustyai_garak::cwe + - id: cwe name: Common Weakness Enumeration description: Common Weakness Enumeration - Software security weaknesses category: security @@ -99,7 +99,7 @@ data: - security - cwe - red_team - - id: trustyai_garak::quick + - id: quick name: Quick Scan description: Quick scan with a single probe for testing and validation category: security