config: Add VLLM_DISABLE_COMPILE_CACHE=1 to environment variables across VLLM scripts.

benchmarks/run_vllm_bench.py

@@ -163,6 +163,7 @@ def run_throughput(model, tp_size, backend_name="Default", output_dir=RESULTS_DI
 
     # ENV Setup: Global + Model Specific
     env = os.environ.copy()
+    env["VLLM_DISABLE_COMPILE_CACHE"] = "1"
     
     # Inject model specific env vars (e.g. for AWQ)
     model_env = MODEL_TABLE[model].get("env", {})