A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 01:00 UTC
[Bugfix] Temporarily disable B200 fp4 MoE layer tests (#40057)
Signed-off-by: Bill Nell <bnell@redhat.com>
bnellnm committed
Apr 16, 2026 at 19:26 UTC
79e799ebbd0a4472f8c6bd846ec676fded664138
1 file changed
+8
tests/kernels/moe/test_moe_layer.py
+8
@@ -465,6 +465,14 @@ def is_valid_config(config: MoETestConfig) -> tuple[bool, str | None]:
465
if config.enable_eplb and config.ep_size == 1:
466
return False, "EPLB only works with EP+DP"
467
468
+ # Disable fp4 tests until flashinfer is updated or the Dockerfile is
469
+ # modified to install cublasLt.h. See #39525.
470
+ if (
471
+ config.quantization == "modelopt_fp4"
472
+ and current_platform.is_device_capability_family(100)
473
+ ):
474
+ return False, "Temporarily skip until #39525 is resolved"
475
+
476
return True, None
477
478