pytorch - 💡(How to fix) Fix torch.compile produces Inf on valid float32 boundary inputs while eager remains finite, leading to catastrophic downstream error propagation [3 comments, 3 participants]

pytorch2026-03-21 14:50:29

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

pytorch/pytorch#178055•Fetched 2026-04-08 01:12:27

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Participants

Timeline (top)

commented ×3labeled ×2closed ×1cross-referenced ×1

Root Cause

input min=1.19828462787679e+29, max=3.4028234663852886e+38 mode cuda+float32+eager: has_nan=False, has_inf=False mode cuda+float32+compile: has_nan=False, has_inf=True The input contains very large values, but they are still legal float32 values (the max is near float32 upper bound / FLT_MAX). So this is not caused by invalid input data.

Fix Action

Fix / Workaround

CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 40 bits physical, 48 bits virtual Byte Order: Little Endian CPU(s): 48 On-line CPU(s) list: 0-47 Vendor ID: GenuineIntel Model name: QEMU Virtual CPU version 2.5+ CPU family: 15 Model: 107 Thread(s) per core: 1 Core(s) per socket: 48 Socket(s): 1 Stepping: 1 BogoMIPS: 4190.15 Flags: fpu de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx lm constant_tsc nopl xtopology cpuid tsc_known_freq pni ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c hypervisor lahf_lm abm cpuid_fault pti bmi1 avx2 bmi2 avx512f avx512dq avx512cd avx512bw avx512vl Hypervisor vendor: KVM Virtualization type: full L1d cache: 1.5 MiB (48 instances) L1i cache: 1.5 MiB (48 instances) L2 cache: 192 MiB (48 instances) L3 cache: 16 MiB (1 instance) NUMA node(s): 1 NUMA node0 CPU(s): 0-47 Vulnerability Gather data sampling: Not affected Vulnerability Itlb multihit: KVM: Mitigation: VMX unsupported 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: Unknown: No mitigations Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Not affected Vulnerability Spec rstack overflow: Not affected 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; BHI Retpoline Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Not affected

Code Example

from pathlib import Path

import torch
import torch.nn as nn
import torch._inductor.config as cfg
BASE = Path(__file__).resolve().parent
WEIGHT = BASE / "sd.pt"
INPUT = BASE / "input.pt"

def make_conv():
    sd = torch.load(str(WEIGHT), map_location="cpu")
    w, b = sd["conv1_1.weight"], sd["conv1_1.bias"]
    m = nn.Conv2d(3, 64, 3, padding=1, bias=True).eval()
    m.weight.data.copy_(w)
    m.bias.data.copy_(b)
    return m

def run(mode: int):
    x = torch.load(str(INPUT), map_location="cpu").float().to("cuda")
    print(f"input min={x.min().item()}, max={x.max().item()}")
    m = make_conv().to("cuda")
    if mode == 1:
        cfg.shape_padding = True
        m = torch.compile(m, dynamic=False)
    with torch.no_grad():
        y = m(x).float()
    tag = "cuda+float32+eager" if mode == 0 else "cuda+float32+compile"
    print(f"mode {tag}: has_nan={bool(torch.isnan(y).any())}, has_inf={bool(torch.isinf(y).any())}}")


if __name__ == "__main__":
    run(0)
    run(1)

---

input min=1.19828462787679e+29, max=3.4028234663852886e+38
mode cuda+float32+eager: has_nan=False, has_inf=False
input min=1.19828462787679e+29, max=3.4028234663852886e+38
mode cuda+float32+compile: has_nan=False, has_inf=True

RAW_BUFFERClick to expand / collapse

🐛 Describe the bug

I observed a reproducible mismatch between eager and compiled execution in a minimal Conv2d test (single/oprat-vgg19-55/test.py) using the same input tensor and the same weights.

Observed logs:

PyTorch documentation notes that edge-case values can lead to slight numerical differences across backends/implementations. However, this case is not a slight inconsistency: the compiled path produces Inf, while eager stays finite on the same data. Once Inf is introduced, it propagates through subsequent layers and can cause catastrophic impact on model behavior (severe output corruption, unstable inference/training).

This appears to be a numerical stability issue in the compile path rather than an acceptable floating-point tolerance difference.

Expected Behavior For valid float32 inputs, compiled execution should remain numerically stable enough to avoid introducing non-finite values (Inf) when eager execution does not.

Actual Behavior Eager mode: finite outputs, no Inf. Compiled mode: Inf appears on the same input/weights. The Inf propagates to downstream layers and can cause catastrophic model-level failures.

from pathlib import Path

import torch
import torch.nn as nn
import torch._inductor.config as cfg
BASE = Path(__file__).resolve().parent
WEIGHT = BASE / "sd.pt"
INPUT = BASE / "input.pt"

def make_conv():
    sd = torch.load(str(WEIGHT), map_location="cpu")
    w, b = sd["conv1_1.weight"], sd["conv1_1.bias"]
    m = nn.Conv2d(3, 64, 3, padding=1, bias=True).eval()
    m.weight.data.copy_(w)
    m.bias.data.copy_(b)
    return m

def run(mode: int):
    x = torch.load(str(INPUT), map_location="cpu").float().to("cuda")
    print(f"input min={x.min().item()}, max={x.max().item()}")
    m = make_conv().to("cuda")
    if mode == 1:
        cfg.shape_padding = True
        m = torch.compile(m, dynamic=False)
    with torch.no_grad():
        y = m(x).float()
    tag = "cuda+float32+eager" if mode == 0 else "cuda+float32+compile"
    print(f"mode {tag}: has_nan={bool(torch.isnan(y).any())}, has_inf={bool(torch.isinf(y).any())}}")


if __name__ == "__main__":
    run(0)
    run(1)

input min=1.19828462787679e+29, max=3.4028234663852886e+38
mode cuda+float32+eager: has_nan=False, has_inf=False
input min=1.19828462787679e+29, max=3.4028234663852886e+38
mode cuda+float32+compile: has_nan=False, has_inf=True

TEST FILE REMOVED

Versions

PyTorch version: 2.10.0+cu126 Is debug build: False CUDA used to build PyTorch: 12.6 ROCm used to build PyTorch: N/A

OS: Ubuntu 24.04.3 LTS (x86_64) GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.39

Python version: 3.10.19 | packaged by conda-forge | (main, Jan 26 2026, 23:45:08) [GCC 14.3.0] (64-bit runtime) Python platform: Linux-6.8.0-90-generic-x86_64-with-glibc2.39 Is CUDA available: True CUDA runtime version: 12.6.20 CUDA_MODULE_LOADING set to: GPU models and configuration: GPU 0: NVIDIA GeForce RTX 3090 GPU 1: NVIDIA GeForce RTX 3090

Nvidia driver version: 560.35.03 cuDNN version: Could not collect Is XPU available: False HIP runtime version: N/A MIOpen runtime version: N/A Is XNNPACK available: True Caching allocator config: N/A

Versions of relevant libraries: [pip3] numpy==1.26.4 [pip3] nvidia-cublas-cu12==12.6.4.1 [pip3] nvidia-cuda-cupti-cu12==12.6.80 [pip3] nvidia-cuda-nvrtc-cu12==12.6.77 [pip3] nvidia-cuda-runtime-cu12==12.6.77 [pip3] nvidia-cudnn-cu12==9.10.2.21 [pip3] nvidia-cufft-cu12==11.3.0.4 [pip3] nvidia-curand-cu12==10.3.7.77 [pip3] nvidia-cusolver-cu12==11.7.1.2 [pip3] nvidia-cusparse-cu12==12.5.4.2 [pip3] nvidia-cusparselt-cu12==0.7.1 [pip3] nvidia-nccl-cu12==2.27.5 [pip3] nvidia-nvjitlink-cu12==12.6.85 [pip3] nvidia-nvtx-cu12==12.6.77 [pip3] onnxruntime-gpu==1.23.2 [pip3] optree==0.18.0 [pip3] pytorch-triton==3.2.0+git4b3bb1f8 [pip3] torch==2.10.0+cu126 [pip3] torchaudio==2.11.0.dev20260127+cu126 [pip3] torchvision==0.25.0+cu126 [pip3] triton==3.6.0+git9844da95 [conda] numpy 1.26.4 pypi_0 pypi [conda] nvidia-cublas-cu12 12.6.4.1 pypi_0 pypi [conda] nvidia-cuda-cupti-cu12 12.6.80 pypi_0 pypi [conda] nvidia-cuda-nvrtc-cu12 12.6.77 pypi_0 pypi [conda] nvidia-cuda-runtime-cu12 12.6.77 pypi_0 pypi [conda] nvidia-cudnn-cu12 9.10.2.21 pypi_0 pypi [conda] nvidia-cufft-cu12 11.3.0.4 pypi_0 pypi [conda] nvidia-curand-cu12 10.3.7.77 pypi_0 pypi [conda] nvidia-cusolver-cu12 11.7.1.2 pypi_0 pypi [conda] nvidia-cusparse-cu12 12.5.4.2 pypi_0 pypi [conda] nvidia-cusparselt-cu12 0.7.1 pypi_0 pypi [conda] nvidia-nccl-cu12 2.27.5 pypi_0 pypi [conda] nvidia-nvjitlink-cu12 12.6.85 pypi_0 pypi [conda] nvidia-nvtx-cu12 12.6.77 pypi_0 pypi [conda] optree 0.18.0 pypi_0 pypi [conda] pytorch-triton 3.2.0+git4b3bb1f8 pypi_0 pypi [conda] torch 2.10.0+cu126 pypi_0 pypi [conda] torchaudio 2.11.0.dev20260127+cu126 pypi_0 pypi [conda] torchvision 0.25.0+cu126 pypi_0 pypi [conda] triton 3.6.0+git9844da95 pypi_0 pypi

extent analysis

Fix Plan

To address the numerical stability issue in the compiled path, we can try the following steps:

Clip input values: Clip the input values to a reasonable range to prevent overflow.
Use a different data type: Use a data type with higher precision, such as float64, to reduce the likelihood of overflow.
Modify the compilation settings: Modify the compilation settings to improve numerical stability.

Here's an example of how you can modify the code to clip the input values:

def run(mode: int):
    x = torch.load(str(INPUT), map_location="cpu").float().to("cuda")
    # Clip input values to a reasonable range
    x = torch.clamp(x, min=-1e30, max=1e30)
    print(f"input min={x.min().item()}, max={x.max().item()}")
    m = make_conv().to("cuda")
    if mode == 1:
        cfg.shape_padding = True
        m = torch.compile(m, dynamic=False)
    with torch.no_grad():
        y = m(x).float()
    tag = "cuda+float32+eager" if mode == 0 else "cuda+float32+compile"
    print(f"mode {tag}: has_nan={bool(torch.isnan(y).any())}, has_inf={bool(torch.isinf(y).any())}")

Alternatively, you can try using a different data type, such as float64:

def run(mode: int):
    x = torch.load(str(INPUT), map_location="cpu").double().to("cuda")  # Use double precision
    print(f"input min={x.min().item()}, max={x.max().item()}")
    m = make_conv().to("cuda")
    if mode == 1:
        cfg.shape_padding = True
        m = torch.compile(m, dynamic=False)
    with torch.no_grad():
        y = m(x).double()  # Use double precision
    tag = "cuda+float64+eager" if mode == 0 else "cuda+float64+compile"
    print(f"mode {tag}: has_nan={bool(torch.isnan(y).any())}, has_inf={bool(torch.isinf(y).any())}")

Verification

To verify that the fix worked, you can run the modified code and check the output for has_inf values. If the issue is resolved, the has_inf value should be False for both eager and compiled modes.

Extra Tips

Be cautious when using float64 as it may impact performance.
Consider using a more robust numerical stability technique, such as using a different activation function or adding noise to the input.
If the issue persists, try modifying the compilation settings or using a different compilation mode.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #vector store #embedding generation #cache error #pipeline error

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

pytorch - 💡(How to fix) Fix torch.compile produces Inf on valid float32 boundary inputs while eager remains finite, leading to catastrophic downstream error propagation [3 comments, 3 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Code Example

🐛 Describe the bug

Versions

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

TRENDING

pytorch - 💡(How to fix) Fix torch.compile produces Inf on valid float32 boundary inputs while eager remains finite, leading to catastrophic downstream error propagation [3 comments, 3 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Code Example

🐛 Describe the bug

Versions

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

RELATED_DISCOVERY

TRENDING