Versions of Relevant Libraries

Oct 13, 2025 by miodeya.com 106 views

Describe the Bug

After compiling a model using the inductor backend, the compiled model outputs significantly different results from the original model. This inconsistency reveals a bug in the compilation process.

Reproducible Script

The following script demonstrates the issue:

import torch

class TestModel(torch.nn.Module):
    def __init__(self):
        super(TestModel, self).__init__()
        self.fc1 = torch.nn.Linear(10, 10)
        self.fc2 = torch.nn.Linear(10, 10)

    def forward(self, x):
        x = self.fc1(x)
        x = torch.nn.functional.gumbel_softmax(x, tau=1.0, hard=False, dim=-1)
        x = self.fc2(x)
        x = torch.nn.functional.softshrink(x, lambd=0.5)
        return x


model = TestModel()
compiled_model = torch.compile(model, backend='inductor')

for i in range(30):
    inputs = torch.randn(1, 10)
    ori_res = model(*inputs)
    compiled_out = compiled_model(*inputs)
    torch.testing.assert_close(ori_res, compiled_out, rtol=1e-3, atol=1e-3)

StackTrace (Inconsistency Results)

The script produces the following error message:

(torch) [qshenaf@sccpu6 0309]$ python torch.nn.functional.gumbel_softmax.py
/data/qshenaf/miniconda3/envs/torch/lib/python3.13/site-packages/torch/_dynamo/guards.py:741: RuntimeWarning: Guards may run slower on Python 3.13.0. Consider upgrading to Python 3.13.1+.
  warnings.warn(
Traceback (most recent call last):
  File "/data/qshenaf/remote_pc/LLM4Converter/bugs/0309/torch.nn.functional.gumbel_softmax.py", line 24, in <module>
    torch.testing.assert_close(ori_res, compiled_out, rtol=1e-3, atol=1e-3)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data/qshenaf/miniconda3/envs/torch/lib/python3.13/site-packages/torch/testing/_comparison.py", line 1519, in assert_close
    raise error_metas[0].to_error(msg)
AssertionError: Tensor-likes are not close!

Mismatched elements: 2 / 10 (20.0%)
Greatest absolute difference: 0.05042219161987305 at index (5,) (up to 0.001 allowed)
Greatest relative difference: inf at index (6,) (up to 0.001 allowed)
(torch) [qshenaf@sccpu6 0309]$

Versions

The following versions of relevant libraries are used:

PyTorch version: 2.7.0.dev20250308+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A
OS: AlmaLinux 9.4 (Seafoam Ocelot) (x86_64)
GCC version: (GCC) 11.4.1 20231218 (Red Hat 11.4.1-3)
Clang version: 17.0.6 (AlmaLinux OS Foundation 17.0.6-5.el9)
CMake version: version 3.26.5
Libc version: glibc-2.34
Python version: 3.13.0 | packaged by Anaconda, Inc. | (main, Oct 7 2024, 21:29:38) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.14.0-427.37.1.el9_4.x86_64-x86_64-with-glibc2.34
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
- GPU 0: NVIDIA GeForce RTX 3090
- GPU 1: NVIDIA GeForce RTX 3090
- GPU 2: NVIDIA GeForce RTX 3090
- GPU 3: NVIDIA GeForce RTX 3090
Nvidia driver version: 560.35.03
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU

The CPU information is as follows:

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 43 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Vendor ID: AuthenticAMD
Model name: AMD Ryzen Threadripper PRO 3975WX 32-Cores
CPU family: 23
Model: 49
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 1
Stepping: 0
Frequency boost: enabled
CPU(s) scaling MHz: 80%
CPU max MHz: 4368.1641
CPU min MHz: 2200.0000
BogoMIPS: 7000.23
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 rapl 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 amd_ppin 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 sev sev_es
Virtualization: AMD-V
L1d cache: 1 MiB (32 instances)
L1i cache: 1 MiB (32 instances)
L2 cache: 16 MiB (32 instances)
L3 cache: 128 MiB (8 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-63
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 Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection
Vulnerability Spec rstack overflow: Mitigation; Safe RET
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; Retpolines, IBPB conditional, STIBP always-on, RSB filling, PBRSB-eIBRS Notaffected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

Versions of Relevant Libraries

The following versions of relevant libraries are used:

numpy==2.2.3
nvidia-cublas-cu12==12.6.4.1
nvidia-cuda-cupti-cu12==12.6.80
nvidia-cuda-nvrtc-cu12==12.6.77
nvidia-cuda-runtime-cu12==12.6.77
nvidia-cudnn-cu12==9.5.1.17
nvidia-cufft-cu12==11.3.0.4
nvidia-curand-cu12==10.3.7.77
nvidia-cusolver-cu12==11.7.1.2
nvidia-cusparse-cu12==12.5.4.2
nvidia-cusparselt-cu12==0
Q&A: Inconsistency Predict Results for Compiled Models with the Original Model ====================================================================================

Q: What is the issue with the compiled model?

A: The compiled model outputs significantly different results from the original model, revealing a bug in the compilation process.

Q: What is the reproducible script that demonstrates the issue?

A: The reproducible script is provided in the previous section. It defines a test model, compiles it using the inductor backend, and then runs a loop to compare the output of the compiled model with the original model.

Q: What is the error message produced by the script?

A: The script produces an error message indicating that the tensor-likes are not close, with a mismatched elements ratio of 20.0% and a greatest absolute difference of 0.05042219161987305 at index (5,).

Q: What are the versions of relevant libraries used in the script?

A: The versions of relevant libraries used in the script are:

PyTorch version: 2.7.0.dev20250308+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A
OS: AlmaLinux 9.4 (Seafoam Ocelot) (x86_64)
GCC version: (GCC) 11.4.1 20231218 (Red Hat 11.4.1-3)
Clang version: 17.0.6 (AlmaLinux OS Foundation 17.0.6-5.el9)
CMake version: version 3.26.5
Libc version: glibc-2.34
Python version: 3.13.0 | packaged by Anaconda, Inc. | (main, Oct 7 2024, 21:29:38) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.14.0-427.37.1.el9_4.x86_64-x86_64-with-glibc2.34
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
- GPU 0: NVIDIA GeForce RTX 3090
- GPU 1: NVIDIA GeForce RTX 3090
- GPU 2: NVIDIA GeForce RTX 3090
- GPU 3: NVIDIA GeForce RTX 3090
Nvidia driver version: 560.35.03
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Q: What is the CPU information?

A: The CPU information is as follows:

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 43 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Vendor ID: AuthenticAMD
Model name: AMD Ryzen Threadripper PRO 3975WX 32-Cores
CPU family: 23
Model: 49
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 1
Stepping: 0
Frequency boost: enabled
CPU(s) scaling MHz: 80%
CPU max MHz: 4368.1641
CPU min MHz: 2200.0000
BogoMIPS: 7000.23
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 rapl 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 amd_ppin 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 sev sev_es
Virtualization: AMD-V
L1d cache: 1 MiB (32 instances)
L1i cache: 1 MiB (32 instances)
L2 cache: 16 MiB (32 instances)
L3 cache: 128 MiB (8 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-63
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 Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection
Vulnerability Spec rstack overflow: Mitigation; Safe RET
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; Retpolines, IBPB conditional, STIBP always-on, RSB filling, PBRSB-eIBRS Notaffected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

Q: What are the versions of relevant libraries used in the script?

A: The versions of relevant libraries used in the script are:

numpy==2.2.3
nvidia-cublas-cu12==12.6.4.1
nvidia-cuda-cupti-cu12==12.6.80
nvidia-cuda-nvrtc-cu12==12.6.77
nvidia-cuda-runtime-cu12==12.6.77
nvidia-cudnn-cu12==9.5.1.17
nvidia-cufft-cu12==11.3.0.4
nvidia-curand-cu12==10.3.7.77
nvidia-cusolver-cu12==11.7.1.2
nvidia-cusparse-cu12==12.5.4.2
nvidia-cusparselt-cu12==0.6.3
nvidia-nccl-cu12==2.25.1
nvidia-nvjitlink-cu12==12.6.85
nvidia-nvtx-cu12==12.6.77
pytorch-triton==3.2.0+git4b3bb1f8
torch==2.7.0.dev20250308+cu126
torchaudio==2.6.0.dev20250308+cu126
torchvision==0.22.0.dev20250308+cu126
triton==3.2.0

Q: What is the next step to resolve the issue?

A: The next step is to investigate the cause of the inconsistency between the compiled model and the original model. This may involve analyzing the compilation process, the model architecture, and the input data to identify the source of the issue.