Revert "Bump FlashInfer to 0.6.15 and revert regressions" (#31625)

2026-07-17 16:46:33 -07:00
parent a01a8e1ed9
commit 304a529558
7 changed files with 15 additions and 20 deletions
@@ -21,7 +21,7 @@ class TestGLM52NVFP4TPMTP(
    model = "nvidia/GLM-5.2-NVFP4"
    tp_size = 4
    mem_fraction_static = 0.8
-    bs_1_speed_thres = 280
+    bs_1_speed_thres = 250
    extra_server_args = [
        "--moe-runner-backend",
        "flashinfer_trtllm",