Fixed dataset download issue and modified buidlkite CI yml to reflect new perf and accuracy script changes

Author: JiriesKaileh

.buildkite/models/meta-llama_Llama-Guard-4-12B.yml

@@ -33,7 +33,7 @@ steps:
       MINIMUM_ACCURACY_THRESHOLD: 0.31
     commands:
       - |
-        .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/test_llama_guard_4_accuracy.sh
+        .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/safety_model_benchmark.sh --mode accuracy
   - label: "Record integration test result for meta-llama/Llama-Guard-4-12B"
     key: "record_meta-llama_Llama-Guard-4-12B_IntegrationTest"
     depends_on: "meta-llama_Llama-Guard-4-12B_IntegrationTest"
@@ -54,7 +54,7 @@ steps:
       queue: tpu_v6e_queue 
     commands:
       - |
-        .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/llama_guard_perf_recipe.sh
+        .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/safety_model_benchmark.sh --mode performance
   
   - label: "Record performance benchmark result for meta-llama/Llama-Guard-4-12B"
     key: "record_meta-llama_Llama-Guard-4-12B_Benchmark"

tests/e2e/benchmarking/safety_model_benchmark.sh

@@ -44,8 +44,8 @@ OUTPUT_LEN_OVERRIDE=20 # Max tokens to generate for safety classification
 # --- DATA PATHS ---
 # Source URL for the AILuminate CSV (Public Raw GitHub Link)
 RAW_CSV_URL="https://raw.githubusercontent.com/mlcommons/ailuminate/main/airr_official_1.0_demo_en_us_prompt_set_release.csv"
-LOCAL_CSV_FILE="/tmp/airr_official_1.0_demo_en_us_prompt_set_release.csv" # Local file path for Pytest
-#GCS_JSONL_URI="gs://jiries/datasets/ailuminate/airr_official_1.0_demo_en_us_prompt_set_release.jsonl" 
+LOCAL_CSV_FILE="/tmp/airr_official_1.0_demo_en_us_prompt_set_release.csv" 
+LOCAL_JSONL_FILE="/tmp/airr_official_1.0_demo_en_us_prompt_set_release.jsonl"
 # ------------------
 
 TEST_MODE=""