Reenable breakable CUDA graph for NemotronH (#34538)

2026-08-13 13:58:10 +08:00
parent ef7208d41d
commit 035c622a14
3 changed files with 3 additions and 16 deletions
@@ -52,6 +52,8 @@ DP_ATTENTION_EP_ARGS = [
    "0.93",
    "--max-prefill-tokens",
    "8192",
+    "--cuda-graph-backend-prefill",
+    "disabled",
 ]

 MTP_ARGS = [