🐛 Describe the bug

I am using opt_einsum package in my model. When I trained my model on different GPUs, the model performance is different. After debugging, I found there is a non-deterministic issue in the torch.einsum function, which is called by opt_einsum package.

My own model, which uses the torch.einsum function, was trained on different GPUs
There is a big difference in the testing AUC between them.
A5000:
testing AUC: 0.6059907834101383

A6000:
testing AUC: 0.6728110599078341

A100:
testing AUC: 0.5714285714285714

This is the simple code to reproduce the non-deterministic issue.

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, TensorDataset
import numpy as np
import os
import opt_einsum as oe
import math
import random

os.putenv("CUBLAS_WORKSPACE_CONFIG", ":4096:8")
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
os.environ["WORLD_SIZE"] = "1"


def setup_seed(seed):
    os.environ['PYTHONHASHSEED'] = str(seed)
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = False
    torch.use_deterministic_algorithms(True)

setup_seed(42)

#np.random.seed(42)
#torch.manual_seed(42)
contract = oe.contract

np.set_printoptions(precision=16)
torch.set_printoptions(precision=16)


class TransposedLinear(nn.Module):
    """ Linear module on the second-to-last dimension """

    def __init__(self, d_input, d_output, bias=True):
        super().__init__()
        print ("--------------TransposedLinear setting---------------")
        print ("d_input=", d_input)
        print ("d_output=", d_output)
        print ("bias=", bias)
        self.weight = nn.Parameter(torch.empty(d_output, d_input))
        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5)) # nn.Linear default init


    def forward(self, x):
        print ("----------------------in--TransposedLinear------------")
        print ("==========before call contract,---------")
        print (x)
        print (x.shape)
        print ("--------weight---------")
        print (self.weight)
        print (self.weight.shape)
        print ("-------------=========-------")
        result = torch.einsum('... u l, v u -> ... v l', x, self.weight)
        print ("==========after call contract,---------")
        print (result)
        print (result.shape)

        return result

# Main function
def main():
   tensor = torch.randn(16, 3072, 64).cuda()

   model = TransposedLinear(3072,3072,True).cuda()
   output = model (tensor)

if __name__ == "__main__":
    main()

Test Environment:
A5000: torch 2.1.2, cuda 12.4
A6000: torch 2.1.2, cuda 11.6
A10: torch 2.4.0, cuda 12.4
A100: torch 2.4.0, cuda 12.4

Attached is the outputs on different GPUs.
a10_output.txt
a5000_output.txt
a100_output.txt
a6000_output.txt

Versions

A100 environment:

Collecting environment information...
PyTorch version: 2.4.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Rocky Linux release 8.10 (Green Obsidian) (x86_64)
GCC version: (GCC) 8.5.0 20210514 (Red Hat 8.5.0-22)
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.28

Python version: 3.9.19 (main, May 6 2024, 19:43:03) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-4.18.0-553.16.1.el8_10.x86_64-x86_64-with-glibc2.28
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA A100-PCIE-40GB
GPU 1: NVIDIA A100-PCIE-40GB
GPU 2: NVIDIA A100-PCIE-40GB
GPU 3: NVIDIA A100-PCIE-40GB

Nvidia driver version: 550.54.15
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
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Thread(s) per core: 2
Core(s) per socket: 16
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 106
Model name: Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz
Stepping: 6
CPU MHz: 2900.000
BogoMIPS: 5800.00
Virtualization: VT-x
L1d cache: 48K
L1i cache: 32K
L2 cache: 1280K
L3 cache: 24576K
NUMA node0 CPU(s): 0-15,32-47
NUMA node1 CPU(s): 16-31,48-63
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 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 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi 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 wbnoinvd dtherm ida arat pln pts hwp_epp avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] pytorch-lightning==2.2.0.post0
[pip3] torch==2.4.0
[pip3] torchaudio==2.4.0
[pip3] torchdata==0.8.0
[pip3] torchlibrosa==0.1.0
[pip3] torchmetrics==1.4.1
[pip3] torchtext==0.18.0
[pip3] torchvision==0.19.0
[pip3] triton==3.0.0
[conda] numpy 1.23.5 pypi_0 pypi
[conda] pytorch-lightning 2.2.0.post0 pypi_0 pypi
[conda] torch 2.4.0 pypi_0 pypi
[conda] torchaudio 2.4.0 pypi_0 pypi
[conda] torchdata 0.8.0 pypi_0 pypi
[conda] torchlibrosa 0.1.0 pypi_0 pypi
[conda] torchmetrics 1.4.1 pypi_0 pypi
[conda] torchtext 0.18.0 pypi_0 pypi
[conda] torchvision 0.19.0 pypi_0 pypi
[conda] triton 3.0.0 pypi_0 pypi

A10 environment:

Collecting environment information...
PyTorch version: 2.4.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Rocky Linux release 8.10 (Green Obsidian) (x86_64)
GCC version: (GCC) 8.5.0 20210514 (Red Hat 8.5.0-22)
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.28

Python version: 3.9.19 (main, May 6 2024, 19:43:03) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-4.18.0-553.16.1.el8_10.x86_64-x86_64-with-glibc2.28
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA A10
Nvidia driver version: 550.54.15
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
Byte Order: Little Endian
CPU(s): 112
On-line CPU(s) list: 0-111
Thread(s) per core: 2
Core(s) per socket: 28
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 106
Model name: Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz
Stepping: 6
CPU MHz: 1166.395
BogoMIPS: 5200.00
Virtualization: VT-x
L1d cache: 48K
L1i cache: 32K
L2 cache: 1280K
L3 cache: 43008K
NUMA node0 CPU(s): 0-27,56-83
NUMA node1 CPU(s): 28-55,84-111
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 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 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi 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 wbnoinvd dtherm ida arat pln pts hwp_epp avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] pytorch-lightning==2.2.0.post0
[pip3] torch==2.4.0
[pip3] torchaudio==2.4.0
[pip3] torchdata==0.8.0
[pip3] torchlibrosa==0.1.0
[pip3] torchmetrics==1.4.1
[pip3] torchtext==0.18.0
[pip3] torchvision==0.19.0
[pip3] triton==3.0.0
[conda] numpy 1.23.5 pypi_0 pypi
[conda] pytorch-lightning 2.2.0.post0 pypi_0 pypi
[conda] torch 2.4.0 pypi_0 pypi
[conda] torchaudio 2.4.0 pypi_0 pypi
[conda] torchdata 0.8.0 pypi_0 pypi
[conda] torchlibrosa 0.1.0 pypi_0 pypi
[conda] torchmetrics 1.4.1 pypi_0 pypi
[conda] torchtext 0.18.0 pypi_0 pypi
[conda] torchvision 0.19.0 pypi_0 pypi
[conda] triton 3.0.0 pypi_0 pypi

A5000 environment:

Collecting environment information...
PyTorch version: 2.1.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.3 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: version 3.27.7
Libc version: glibc-2.31

Python version: 3.9.12 (main, Apr 5 2022, 06:56:58) [GCC 7.5.0] (64-bit runtime)
Python platform: Linux-5.13.0-48-generic-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA RTX A5000
GPU 1: NVIDIA RTX A5000

Nvidia driver version: 535.161.07
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.4.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.4.1
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
Byte Order: Little Endian
Address sizes: 43 bits physical, 48 bits virtual
CPU(s): 64
On-line CPU(s) list: 0-63
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 1
NUMA node(s): 2
Vendor ID: AuthenticAMD
CPU family: 23
Model: 49
Model name: AMD Ryzen Threadripper PRO 3975WX 32-Cores
Stepping: 0
Frequency boost: enabled
CPU MHz: 2200.000
CPU max MHz: 4368.1641
CPU min MHz: 2200.0000
BogoMIPS: 6986.94
Virtualization: AMD-V
L1d cache: 1 MiB
L1i cache: 1 MiB
L2 cache: 16 MiB
L3 cache: 128 MiB
NUMA node0 CPU(s): 0-15,32-47
NUMA node1 CPU(s): 16-31,48-63
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, STIBP conditional, RSB filling
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sme sev sev_es

Versions of relevant libraries:
[pip3] flake8==3.9.2
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.22.4
[pip3] numpydoc==1.2
[pip3] onnx==1.13.1
[pip3] onnxruntime==1.14.1
[pip3] open-clip-torch==2.7.0
[pip3] pytorch-lightning==1.7.7
[pip3] torch==2.1.0
[pip3] torchaudio==2.1.0
[pip3] torchdata==0.7.0
[pip3] torchdiffeq==0.2.3
[pip3] torchmetrics==0.11.4
[pip3] torchsde==0.2.5
[pip3] torchtext==0.16.0
[pip3] torchvision==0.16.0
[pip3] triton==2.1.0
[conda] blas 1.0 mkl
[conda] mkl 2021.4.0 h06a4308_640
[conda] mkl-service 2.4.0 py39h7f8727e_0
[conda] mkl_fft 1.3.1 py39hd3c417c_0
[conda] mkl_random 1.2.2 py39h51133e4_0
[conda] numpy 1.22.4 pypi_0 pypi
[conda] numpydoc 1.2 pyhd3eb1b0_0
[conda] open-clip-torch 2.7.0 pypi_0 pypi
[conda] pytorch-lightning 1.7.7 pypi_0 pypi
[conda] torch 2.1.0 pypi_0 pypi
[conda] torchaudio 2.1.0 pypi_0 pypi
[conda] torchdata 0.7.0 pypi_0 pypi
[conda] torchdiffeq 0.2.3 pypi_0 pypi
[conda] torchmetrics 0.11.4 pypi_0 pypi
[conda] torchsde 0.2.5 pypi_0 pypi
[conda] torchtext 0.16.0 pypi_0 pypi
[conda] torchvision 0.16.0 pypi_0 pypi
[conda] triton 2.1.0 pypi_0 pypi

A6000 environment:

Collecting environment information...
PyTorch version: 2.1.2+cu118
Is debug build: False
CUDA used to build PyTorch: 11.8
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.4 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: version 3.27.5
Libc version: glibc-2.31

Python version: 3.9.18 (main, Sep 11 2023, 13:41:44) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.13.0-51-generic-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 11.6.124
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000

Nvidia driver version: 510.73.05
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
Byte Order: Little Endian
Address sizes: 43 bits physical, 48 bits virtual
CPU(s): 64
On-line CPU(s) list: 0-63
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 1
NUMA node(s): 2
Vendor ID: AuthenticAMD
CPU family: 23
Model: 49
Model name: AMD Ryzen Threadripper PRO 3975WX 32-Cores
Stepping: 0
Frequency boost: enabled
CPU MHz: 2200.000
CPU max MHz: 4368.1641
CPU min MHz: 2200.0000
BogoMIPS: 6986.61
Virtualization: AMD-V
L1d cache: 1 MiB
L1i cache: 1 MiB
L2 cache: 16 MiB
L3 cache: 128 MiB
NUMA node0 CPU(s): 0-15,32-47
NUMA node1 CPU(s): 16-31,48-63
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, STIBP conditional, RSB filling
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sme sev sev_es

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] pytorch-lightning==2.0.9
[pip3] pytorch-triton==2.1.0+6e4932cda8
[pip3] torch==2.1.2+cu118
[pip3] torchaudio==2.1.2+cu118
[pip3] torchmetrics==1.2.0
[pip3] torchvision==0.16.2+cu118
[pip3] triton==2.1.0
[conda] numpy 1.23.5 pypi_0 pypi
[conda] pytorch-lightning 2.0.9 pypi_0 pypi
[conda] pytorch-triton 2.1.0+6e4932cda8 pypi_0 pypi
[conda] torch 2.1.2+cu118 pypi_0 pypi
[conda] torchaudio 2.1.2+cu118 pypi_0 pypi
[conda] torchmetrics 1.2.0 pypi_0 pypi
[conda] torchvision 0.16.2+cu118 pypi_0 pypi
[conda] triton 2.1.0 pypi_0 pypi

cc @csarofeen @ptrblck @xwang233 @eqy @mruberry @kurtamohler