vllm icon indicating copy to clipboard operation
vllm copied to clipboard
verified publisher icon vllm-project

[Bug]: Can't run Flashinfer MoE TRTLLM backend FP4 for Qwen3 235B

Open Victor49152 opened this issue 1 month ago • 3 comments

Your current environment

The output of python collect_env.py 
Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 24.04.1 LTS (x86_64)
GCC version                  : (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version                : Could not collect
CMake version                : version 3.31.4
Libc version                 : glibc-2.39

==============================
       PyTorch Info
==============================
PyTorch version              : 2.9.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.3 (main, Jan 17 2025, 18:03:48) [GCC 13.3.0] (64-bit runtime)
Python platform              : Linux-6.8.0-1032-nvidia-x86_64-with-glibc2.39

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.61
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : 
GPU 0: NVIDIA B200
GPU 1: NVIDIA B200
GPU 2: NVIDIA B200
GPU 3: NVIDIA B200
GPU 4: NVIDIA B200
GPU 5: NVIDIA B200
GPU 6: NVIDIA B200
GPU 7: NVIDIA B200

Nvidia driver version        : 580.65.06
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.7.1
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Address sizes:                        52 bits physical, 57 bits virtual
Byte Order:                           Little Endian
CPU(s):                               224
On-line CPU(s) list:                  0-223
Vendor ID:                            GenuineIntel
Model name:                           INTEL(R) XEON(R) PLATINUM 8570
CPU family:                           6
Model:                                207
Thread(s) per core:                   2
Core(s) per socket:                   56
Socket(s):                            2
Stepping:                             2
CPU(s) scaling MHz:                   31%
CPU max MHz:                          4000.0000
CPU min MHz:                          800.0000
BogoMIPS:                             4200.00
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                       VT-x
L1d cache:                            5.3 MiB (112 instances)
L1i cache:                            3.5 MiB (112 instances)
L2 cache:                             224 MiB (112 instances)
L3 cache:                             600 MiB (2 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-55,112-167
NUMA node1 CPU(s):                    56-111,168-223
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Not affected
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

==============================
Versions of relevant libraries
==============================
[pip3] flashinfer-python==0.4.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.15.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.2.1
[pip3] nvidia-dali-cuda120==1.46.0
[pip3] nvidia-ml-py==13.580.82
[pip3] nvidia-modelopt==0.23.0
[pip3] nvidia-modelopt-core==0.23.0
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvcomp-cu12==4.1.0.6
[pip3] nvidia-nvimgcodec-cu12==0.4.1.21
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvjpeg2k-cu12==0.8.1.40
[pip3] nvidia-nvshmem-cu12==3.3.20
[pip3] nvidia-nvtiff-cu12==0.4.0.62
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] nvidia-pyindex==1.0.9
[pip3] onnx==1.17.0
[pip3] optree==0.14.0
[pip3] pynvml==11.4.1
[pip3] pytorch-triton==3.2.0+git0d4682f0b.nvinternal
[pip3] pyzmq==26.2.1
[pip3] torch==2.9.0
[pip3] torch_geometric==2.5.3
[pip3] torch_tensorrt==2.6.0a0
[pip3] torchaudio==2.9.0
[pip3] torchprofile==0.0.4
[pip3] torchvision==0.24.0
[pip3] transformers==4.57.1
[pip3] triton==3.5.0
[pip3] triton_kernels==1.0.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.11.1rc3
vLLM Build Flags:
  CUDA Archs: 7.5 8.0 8.6 9.0 10.0 12.0+PTX; ROCm: Disabled
GPU Topology:
  	GPU0	GPU1	GPU2	GPU3	GPU4	GPU5	GPU6	GPU7	NIC0	NIC1	NIC2	NIC3	NIC4	NIC5	NIC6	NIC7	NIC8	NIC9	NIC10	NIC11	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	NV18	NV18	NV18	NV18	NV18	NV18	NV18	PXB	NODE	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	0-55,112-167	0		N/A
GPU1	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NV18	NODE	NODE	NODE	PXB	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	0-55,112-167	0		N/A
GPU2	NV18	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NODE	NODE	NODE	NODE	PXB	NODE	SYS	SYS	SYS	SYS	SYS	SYS	0-55,112-167	0		N/A
GPU3	NV18	NV18	NV18	 X 	NV18	NV18	NV18	NV18	NODE	NODE	NODE	NODE	NODE	PXB	SYS	SYS	SYS	SYS	SYS	SYS	0-55,112-167	0		N/A
GPU4	NV18	NV18	NV18	NV18	 X 	NV18	NV18	NV18	SYS	SYS	SYS	SYS	SYS	SYS	PXB	NODE	NODE	NODE	NODE	NODE	56-111,168-223	1		N/A
GPU5	NV18	NV18	NV18	NV18	NV18	 X 	NV18	NV18	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	PXB	NODE	NODE	56-111,168-223	1		N/A
GPU6	NV18	NV18	NV18	NV18	NV18	NV18	 X 	NV18	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	PXB	NODE	56-111,168-223	1		N/A
GPU7	NV18	NV18	NV18	NV18	NV18	NV18	NV18	 X 	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	NODE	PXB	56-111,168-223	1		N/A
NIC0	PXB	NODE	NODE	NODE	SYS	SYS	SYS	SYS	 X 	NODE	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS				
NIC1	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	NODE	 X 	PIX	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS				
NIC2	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	NODE	PIX	 X 	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS				
NIC3	NODE	PXB	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	NODE	 X 	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS				
NIC4	NODE	NODE	PXB	NODE	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	 X 	NODE	SYS	SYS	SYS	SYS	SYS	SYS				
NIC5	NODE	NODE	NODE	PXB	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	NODE	 X 	SYS	SYS	SYS	SYS	SYS	SYS				
NIC6	SYS	SYS	SYS	SYS	PXB	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	 X 	NODE	NODE	NODE	NODE	NODE				
NIC7	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	NODE	 X 	PIX	NODE	NODE	NODE				
NIC8	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	NODE	PIX	 X 	NODE	NODE	NODE				
NIC9	SYS	SYS	SYS	SYS	NODE	PXB	NODE	NODE	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	 X 	NODE	NODE				
NIC10	SYS	SYS	SYS	SYS	NODE	NODE	PXB	NODE	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	 X 	NODE				
NIC11	SYS	SYS	SYS	SYS	NODE	NODE	NODE	PXB	SYS	SYS	SYS	SYS	SYS	SYS	NODE	NODE	NODE	NODE	NODE	 X 				

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

NIC Legend:

  NIC0: mlx5_4
  NIC1: mlx5_5
  NIC2: mlx5_6
  NIC3: mlx5_7
  NIC4: mlx5_8
  NIC5: mlx5_9
  NIC6: mlx5_10
  NIC7: mlx5_11
  NIC8: mlx5_12
  NIC9: mlx5_13
  NIC10: mlx5_14
  NIC11: mlx5_15

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=void
CUBLAS_VERSION=12.8.3.14
NVIDIA_REQUIRE_CUDA=cuda>=9.0
CUDA_CACHE_DISABLE=1
TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0+PTX
NCCL_VERSION=2.25.1
NVIDIA_DRIVER_CAPABILITIES=compute,utility,video
TORCH_NCCL_USE_COMM_NONBLOCKING=0
NVIDIA_PRODUCT_NAME=PyTorch
CUDA_VERSION=12.8.0.038
PYTORCH_VERSION=2.7.0a0+ecf3bae
PYTORCH_BUILD_NUMBER=0
VLLM_ATTENTION_BACKEND=FLASHINFER
CUDNN_FRONTEND_VERSION=1.10.0
CUDNN_VERSION=9.7.1.26
VLLM_FLASHINFER_MOE_BACKEND=throughput
PYTORCH_HOME=/opt/pytorch/pytorch
LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/torch/lib:/usr/local/lib/python3.12/dist-packages/torch_tensorrt/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NVIDIA_BUILD_ID=143088496
NVIDIA_CTK_LIBCUDA_DIR=/usr/lib/x86_64-linux-gnu
CUDA_DRIVER_VERSION=570.86.10
PYTORCH_BUILD_VERSION=2.7.0a0+ecf3bae
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
CUDA_MODULE_LOADING=LAZY
NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS=
VLLM_USE_FLASHINFER_MOE_FP4=1
NVIDIA_PYTORCH_VERSION=25.02
TORCH_ALLOW_TF32_CUBLAS_OVERRIDE=1
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1

=================================
[9ec78181b8c7:56683:0:56683] Caught signal 8 (Floating point exception: integer divide by zero)
==== backtrace (tid:  56683) ====
 0  /opt/hpcx/ucx/lib/libucs.so.0(ucs_handle_error+0x2e4) [0x7d74482f5654]
 1  /opt/hpcx/ucx/lib/libucs.so.0(+0x3684c) [0x7d74482f584c]
 2  /opt/hpcx/ucx/lib/libucs.so.0(+0x36bda) [0x7d74482f5bda]
 3  /usr/lib/x86_64-linux-gnu/libc.so.6(+0x45330) [0x7d78508a0330]
 4  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(_ZN3moe3dev7routing15routingDeepSeek12KernelParamsIf13__nv_bfloat16Lb0ELb1EE15setKernelParamsERKNS2_4DataE+0x7f) [0x7d7281d7d39f]
 5  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(_ZN3moe3dev7routing15routingDeepSeek7runImplERNS2_4DataEPv+0x358d) [0x7d7281d6cdad]
 6  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(_ZN12tensorrt_llm7kernels13trtllmgen_moe7Routing6Runner3runEPvS4_iiiiiiifPiS5_S5_S5_S5_S5_S4_S5_S5_S5_S5_N11batchedGemm6trtllm3gen5DtypeEbbNS2_17RoutingMethodTypeEP11CUstream_st+0x1e81) [0x7d7281d66791]
 7  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(_ZN10flashinfer35trtllm_fp4_block_scale_moe_launcherEN3tvm3ffi8OptionalINS1_10TensorViewEvEES3_S3_S4_S3_S4_S3_S3_S4_S4_S4_S4_S3_S3_S4_S4_S4_S4_llNS2_IlvEES5_lllNS2_IdvEEllbRN12tensorrt_llm7kernels13trtllmgen_moe3MoE6RunnerEN11batchedGemm6trtllm3gen5DtypeESG_lbS3_+0x20b9) [0x7d7281ceabc9]
 8  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(_ZN10flashinfer26trtllm_fp4_block_scale_moeEN3tvm3ffi8OptionalINS1_10TensorViewEvEES3_S3_S4_S3_S4_S3_S3_S4_S4_S4_S4_S3_S3_S4_S4_S4_S4_llNS2_IlvEES5_lllNS2_IdvEEllbblS3_l+0xec2) [0x7d7281cfa652]
 9  /root/.cache/flashinfer/0.4.1/100a/cached_ops/fused_moe_trtllm_sm100/fused_moe_trtllm_sm100.so(__tvm_ffi_trtllm_fp4_block_scale_moe+0x1e5da) [0x7d7281d5b0fa]
10  /usr/local/lib/python3.12/dist-packages/tvm_ffi/lib/libtvm_ffi.so(+0xb460f) [0x7d76a89e160f]
11  /usr/local/lib/python3.12/dist-packages/tvm_ffi/lib/libtvm_ffi.so(+0xb3f7c) [0x7d76a89e0f7c]
12  /usr/local/lib/python3.12/dist-packages/tvm_ffi/core.abi3.so(+0x49ea3) [0x7d76a8914ea3]
13  VLLM::Worker_TP6(PyObject_Call+0x6c) [0x54b3ac]
14  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
15  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x85294a) [0x7d7842b9094a]
16  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0xb584ab) [0x7d7842e964ab]
17  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_cpu.so(+0x5b4222d) [0x7d78339c822d]
18  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(_ZN5torch3jit24invokeOperatorFromPythonERKSt6vectorISt10shared_ptrINS0_8OperatorEESaIS4_EERKN8pybind114argsERKNS9_6kwargsESt8optionalIN3c1011DispatchKeyEE+0x394) [0x7d7842c49a04]
19  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(_ZN5torch3jit37_get_operation_for_overload_or_packetERKSt6vectorISt10shared_ptrINS0_8OperatorEESaIS4_EEN3c106SymbolERKN8pybind114argsERKNSB_6kwargsEbSt8optionalINS9_11DispatchKeyEE+0x1a9) [0x7d7842c49dc9]
20  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x81539c) [0x7d7842b5339c]
21  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x81584f) [0x7d7842b5384f]
22  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x3767d2) [0x7d78426b47d2]
23  VLLM::Worker_TP6() [0x58238f]
24  VLLM::Worker_TP6(PyObject_Call+0x6c) [0x54b3ac]
25  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
26  VLLM::Worker_TP6(_PyObject_Call_Prepend+0xc2) [0x54aa72]
27  VLLM::Worker_TP6() [0x5a3b08]
28  VLLM::Worker_TP6(_PyObject_MakeTpCall+0x75) [0x549225]
29  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0xa89) [0x5d7909]
30  VLLM::Worker_TP6(_PyObject_Call_Prepend+0xc2) [0x54aa72]
31  VLLM::Worker_TP6() [0x5a3b08]
32  VLLM::Worker_TP6(_PyObject_MakeTpCall+0x75) [0x549225]
33  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0xa89) [0x5d7909]
34  VLLM::Worker_TP6(_PyObject_Call_Prepend+0xc2) [0x54aa72]
35  VLLM::Worker_TP6() [0x5a3b08]
36  VLLM::Worker_TP6(PyObject_Call+0x6c) [0x54b3ac]
37  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
38  VLLM::Worker_TP6(_PyObject_Call_Prepend+0xc2) [0x54aa72]
39  VLLM::Worker_TP6() [0x5a3b08]
40  VLLM::Worker_TP6(_PyObject_MakeTpCall+0x75) [0x549225]
41  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0xa89) [0x5d7909]
42  VLLM::Worker_TP6() [0x54cd7d]
43  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
44  VLLM::Worker_TP6() [0x54cd7d]
45  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
46  VLLM::Worker_TP6() [0x54cd7d]
47  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
48  VLLM::Worker_TP6(_PyObject_Call_Prepend+0xc2) [0x54aa72]
49  VLLM::Worker_TP6() [0x5a3b08]
50  VLLM::Worker_TP6(PyObject_Call+0x6c) [0x54b3ac]
51  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
52  VLLM::Worker_TP6() [0x54cd7d]
53  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
54  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x6f6a45) [0x7d7842a34a45]
55  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
56  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x6f6aa0) [0x7d7842a34aa0]
57  VLLM::Worker_TP6() [0x54cd7d]
58  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
59  /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_python.so(+0x6f6aa0) [0x7d7842a34aa0]
60  VLLM::Worker_TP6() [0x54cd7d]
61  VLLM::Worker_TP6(_PyEval_EvalFrameDefault+0x4b83) [0x5dba03]
=================================

🐛 Describe the bug

The integer divided by zero error comes from these two lines.

To reproduce the issue, set the environment variables

export VLLM_ATTENTION_BACKEND=FLASHINFER
export VLLM_USE_FLASHINFER_MOE_FP4=1 
export VLLM_FLASHINFER_MOE_BACKEND=latency

and then run vllm serve nvidia/Qwen3-235B-A22B-FP4 --tensor-parallel-size 8 --mm-encoder-tp-mode data --async-scheduling --enable-expert-parallel

Before submitting a new issue...

  • [x] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

Victor49152 avatar Nov 03 '25 23:11 Victor49152

  1. The integer divided by zero issue is resolved by switching routing method to "Renormalize"
  2. But by simply switching to this routing method, the accuracy of model is seriously damaged, on gms8k, the original ckpt nvidia/Qwen3-235B-A22B-FP4 use flashinfer cutlass kernel reaches 0.4397 accuracy (pretty bad though), but switching to flashinfer.trtllm kernel, the accuracy drops to zero

Victor49152 avatar Nov 05 '25 23:11 Victor49152

This issue will be fixed by #28569 . cc @jiahanc

wangshangsam avatar Nov 13 '25 21:11 wangshangsam

root causes of the issue

  1. The flashinfer FP4 TRTLLM-GEN MOE originally only support 2 routing methods: DeepSeek_V3 and Llama4, so the routing is hardcoded to these 2. After https://github.com/vllm-project/vllm/pull/27492 multiple routing methods are supported, it should derive from MOE layer.
  2. flashinfer FP4 TRTLLM-GEN MOE requires global_sf same with flashinfer cutlass. It should be added to https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/quantization/utils/flashinfer_utils.py#L273-L276, but somehow missed. W/O global_sf, the FP4 quant will use the 1st element across all experts, which causes quantize accuracy issues. Both these 2 are fixed in https://github.com/vllm-project/vllm/pull/28569

jiahanc avatar Nov 14 '25 00:11 jiahanc