Update platform_agnostic_inference_code_score21.py

platform_agnostic_inference_code_score21.py

@@ -177,7 +177,7 @@ def load_model():
         "--max-num-seqs=11",
         "--max-cudagraph-capture-size=2048",
         "--gpu-memory-utilization=0.95",
-        "--kv-cache-dtype=auto",
+        "--kv-cache-dtype=fp8_e4m3",
         "--stream-interval=200",
         "--enable-prefix-caching",
         "--uvicorn-log-level debug",