vllm: [Bug]: "Triton Error [CUDA]: device kernel image is invalid" when loading Mixtral-8x7B-Instruct-v0.1 in fused_moe.py

Your current environment

2024-06-19 17:30:02,514 - [Collecting environment information...
PyTorch version: 2.3.0
Is debug build: False
CUDA used to build PyTorch: 11.8
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.29.5
Libc version: glibc-2.35

Python version: 3.10.14 (main, Mar 21 2024, 16:24:04) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.4.231-137.341.amzn2.x86_64-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA A100-SXM4-40GB
GPU 1: NVIDIA A100-SXM4-40GB
GPU 2: NVIDIA A100-SXM4-40GB
GPU 3: NVIDIA A100-SXM4-40GB

Nvidia driver version: 470.161.03
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                    x86_64
CPU op-mode(s):                  32-bit, 64-bit
Address sizes:                   46 bits physical, 48 bits virtual
Byte Order:                      Little Endian
CPU(s):                          96
On-line CPU(s) list:             0-95
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz
CPU family:                      6
Model:                           85
Thread(s) per core:              2
Core(s) per socket:              24
Socket(s):                       2
Stepping:                        7
BogoMIPS:                        5999.99
Flags:                           fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single pti fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves ida arat pku ospke
Hypervisor vendor:               KVM
Virtualization type:             full
L1d cache:                       1.5 MiB (48 instances)
L1i cache:                       1.5 MiB (48 instances)
L2 cache:                        48 MiB (48 instances)
L3 cache:                        71.5 MiB (2 instances)
NUMA node(s):                    2
NUMA node0 CPU(s):               0-23,48-71
NUMA node1 CPU(s):               24-47,72-95
Vulnerability Itlb multihit:     KVM: Vulnerable
Vulnerability L1tf:              Mitigation; PTE Inversion
Vulnerability Mds:               Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
Vulnerability Meltdown:          Mitigation; PTI
Vulnerability Mmio stale data:   Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
Vulnerability Retbleed:          Vulnerable
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1:        Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:        Mitigation; Retpolines, STIBP disabled, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds:             Not affected
Vulnerability Tsx async abort:   Not affected

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] optree==0.11.0
[pip3] torch==2.3.0
[pip3] torchaudio==2.3.0
[pip3] torchelastic==0.2.2
[pip3] torchvision==0.18.0
[pip3] transformers==4.41.2
[pip3] triton==2.3.0
[conda] blas                      1.0                         mkl  
[conda] ffmpeg                    4.3                  hf484d3e_0    pytorch
[conda] libjpeg-turbo             2.0.0                h9bf148f_0    pytorch
[conda] mkl                       2023.1.0         h213fc3f_46344  
[conda] mkl-service               2.4.0           py310h5eee18b_1  
[conda] mkl_fft                   1.3.8           py310h5eee18b_0  
[conda] mkl_random                1.2.4           py310hdb19cb5_0  
[conda] numpy                     1.26.4          py310h5f9d8c6_0  
[conda] numpy-base                1.26.4          py310hb5e798b_0  
[conda] optree                    0.11.0                   pypi_0    pypi
[conda] pytorch                   2.3.0           py3.10_cuda11.8_cudnn8.7.0_0    pytorch
[conda] pytorch-cuda              11.8                 h7e8668a_5    pytorch
[conda] pytorch-mutex             1.0                        cuda    pytorch
[conda] torchaudio                2.3.0               py310_cu118    pytorch
[conda] torchelastic              0.2.2                    pypi_0    pypi
[conda] torchtriton               2.3.0                     py310    pytorch
[conda] torchvision               0.18.0              py310_cu118    pytorch
[conda] transformers              4.41.2                   pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.0.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    GPU2    GPU3    CPU Affinity    NUMA Affinity
GPU0     X      NV12    NV12    NV12    0-23,48-71      0
GPU1    NV12     X      NV12    NV12    0-23,48-71      0
GPU2    NV12    NV12     X      NV12    24-47,72-95     1
GPU3    NV12    NV12    NV12     X      24-47,72-95     1

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

🐛 Describe the bug

pip3 install https://github.com/vllm-project/vllm/releases/download/v0.5.0.post1/vllm-0.5.0.post1+cu118-cp310-cp310-manylinux1_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118

I hit this issue when running vllm worker in fastchat: https://github.com/lm-sys/FastChat/blob/main/fastchat/serve/vllm_worker.py
 
engine = AsyncLLMEngine.from_engine_args(engine_args)

it failed when loading the model `Mixtral-8x7B-Instruct-v0.1` 

Error:

INFO 06-20 00:31:58 model_runner.py:160] Loading model weights took 21.7573 GB
INFO 06-20 00:31:58 model_runner.py:160] Loading model weights took 21.7573 GB
INFO 06-20 00:31:58 model_runner.py:160] Loading model weights took 21.7573 GB
INFO 06-20 00:31:58 model_runner.py:160] Loading model weights took 21.7573 GB
INFO 06-20 00:32:01 fused_moe.py:300] Using configuration from /opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json for MoE layer.
INFO 06-20 00:32:01 fused_moe.py:300] Using configuration from /opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json for MoE layer.
INFO 06-20 00:32:01 fused_moe.py:300] Using configuration from /opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json for MoE layer.
INFO 06-20 00:32:01 fused_moe.py:300] Using configuration from /opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json for MoE layer.
wait for model worker to start
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226] Exception in worker VllmWorkerProcess while processing method determine_num_available_blocks: Triton Error [CUDA]: device kernel image is invalid, Traceback (most recent call last):
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/executor/multiproc_worker_utils.py", line 223, in _run_worker_process
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     output = executor(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return func(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/worker/worker.py", line 162, in determine_num_available_blocks
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     self.model_runner.profile_run()
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return func(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/worker/model_runner.py", line 844, in profile_run
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     self.execute_model(seqs, kv_caches)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return func(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/worker/model_runner.py", line 749, in execute_model
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     hidden_states = model_executable(
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return self._call_impl(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return forward_call(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/models/mixtral.py", line 535, in forward
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     hidden_states = self.model(input_ids, positions, kv_caches,
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return self._call_impl(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return forward_call(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/models/mixtral.py", line 468, in forward
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     hidden_states, residual = layer(positions, hidden_states,
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return self._call_impl(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return forward_call(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/models/mixtral.py", line 424, in forward
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     hidden_states = self.block_sparse_moe(hidden_states)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return self._call_impl(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return forward_call(*args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/models/mixtral.py", line 271, in forward
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     final_hidden_states = fused_moe(hidden_states,
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/fused_moe.py", line 515, in fused_moe
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return fused_experts(hidden_states,
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/fused_moe.py", line 427, in fused_experts
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     invoke_fused_moe_kernel(hidden_states,
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/vllm/model_executor/layers/fused_moe/fused_moe.py", line 245, in invoke_fused_moe_kernel
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     fused_moe_kernel[grid](
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/triton/runtime/jit.py", line 167, in <lambda>
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/triton/runtime/jit.py", line 425, in run
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     kernel.run(grid_0, grid_1, grid_2, kernel.num_warps, kernel.num_ctas,  # number of warps/ctas per instance
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/triton/compiler/compiler.py", line 255, in __getattribute__
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     self._init_handles()
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]   File "/opt/conda/lib/python3.10/site-packages/triton/compiler/compiler.py", line 250, in _init_handles
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226]     self.module, self.function, self.n_regs, self.n_spills = driver.utils.load_binary(
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226] RuntimeError: Triton Error [CUDA]: device kernel image is invalid
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226] 
ERROR 06-20 00:32:02 multiproc_worker_utils.py:226] Exception in worker VllmWorkerProcess while processing method determine_num_available_blocks: Triton Error [CUDA]: device kernel image is invalid, Traceback (most recent call last):

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 18 (6 by maintainers)

Most upvoted comments

把你自己的cuda 下面/usr/local/cuda/bin/ptxas 覆盖 triton 目录下的ptxas就好啦!就是这么简单!

把你自己的cuda 下面/usr/local/cuda/bin/ptxas 覆盖 triton 目录下的ptxas就好啦!就是这么简单!

Can you please explain in a little more detail what needs to be done?

In my case, I copied the ptxas file from /usr/local/cuda/bin/ (Linux) and replaced the one in the directory .../python3.8/site-packages/triton/third_party/cuda/bin (specific to my environment).

The ptxas file is used by nvcc, and the error message indicated that the existing ptxas version was outdated.

In my case, I copied the ptxas file from /usr/local/cuda/bin/ (Linux) and replaced the one in the directory .../python3.8/site-packages/triton/third_party/cuda/bin (specific to my environment).

The ptxas file is used by nvcc, and the error message indicated that the existing ptxas version was outdated.

good idea, thx! i did this and it works! but my path is different from yours. i copied ptxas file to .../lib/python3.12/site-packages/triton/backends/nvidia/bin/.

For anyone else facing the same issue: For us this was resolved when we updated our CUDA version from 11.8 to 12.2

Is there any other solutions except updating the driver version? 😢 We can only use driver 470.x and cu118 in our cluster😭