🐛 Describe the bug

The zero_infinity parameter in torch.nn.functional.ctc_loss currently has no effect when the cuDNN backend is used. For example, the following raises an AssertionError, even though zero_infinity=True:

import torch
probs = torch.nn.functional.one_hot(torch.tensor([0], device='cuda'), num_classes=2).float()
log_probs = torch.log(probs).unsqueeze(1).requires_grad_()
targets = torch.tensor([1], device='cuda', dtype=torch.int32)
input_lengths = torch.tensor([1], device='cuda', dtype=torch.int32)
target_lengths = torch.tensor([1], device='cuda', dtype=torch.int32)
loss = torch.nn.functional.ctc_loss(
    log_probs, targets, input_lengths, target_lengths, reduction='sum', zero_infinity=True
)
assert torch.isfinite(loss)

A similar issue applies to the gradients.

I believe this issue may have gone unnoticed for some time due to cuDNN's unusual behavior for inputs that should cause a divergence. For instance, the following does not raise an AssertionError (though zero_infinity=False):

import torch
device = 'cuda'
log_probs = torch.nn.functional.log_softmax(torch.randn(2, 1, 3, device=device), dim=-1)
targets = torch.tensor([0, 0], device=device, dtype=torch.int32)
input_lengths = torch.tensor([2], device=device, dtype=torch.int32)
target_lengths = torch.tensor([2], device=device, dtype=torch.int32)
loss = torch.nn.functional.ctc_loss(
    log_probs, targets, input_lengths, target_lengths, reduction='sum', zero_infinity=False
)
assert torch.isfinite(loss)

This should be compared with the case device = 'cpu' (or in which cuDNN is disabled).

I have a fix ready and will open an MR shortly.

Versions

Collecting environment information... PyTorch version: 2.12.0a0+gitaee4170 Is debug build: False CUDA used to build PyTorch: 13.0 ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04.4 LTS (x86_64) GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 Clang version: Could not collect CMake version: version 4.2.3 Libc version: glibc-2.39

Python version: 3.11.14 (main, Feb 12 2026, 00:42:50) [Clang 21.1.4 ] (64-bit runtime) Python platform: Linux-6.8.0-1046-nvidia-x86_64-with-glibc2.39 Is CUDA available: True CUDA runtime version: 13.0.88 CUDA_MODULE_LOADING set to: GPU models and configuration: GPU 0: NVIDIA A100-SXM4-40GB Nvidia driver version: 580.126.20 cuDNN version: Probably one of the following: /usr/lib/x86_64-linux-gnu/libcudnn.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.19.1 /usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.19.1 Is XPU available: False HIP runtime version: N/A MIOpen runtime version: N/A Is XNNPACK available: False Caching allocator config: N/A

CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 48 bits physical, 48 bits virtual Byte Order: Little Endian CPU(s): 30 On-line CPU(s) list: 0-29 Vendor ID: AuthenticAMD Model name: AMD EPYC 7J13 64-Core Processor CPU family: 25 Model: 1 Thread(s) per core: 1 Core(s) per socket: 1 Socket(s): 30 Stepping: 1 BogoMIPS: 4899.99 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy svm cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw perfctr_core ssbd ibrs ibpb stibp vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves clzero xsaveerptr wbnoinvd arat npt nrip_save umip pku ospke vaes vpclmulqdq rdpid fsrm arch_capabilities Virtualization: AMD-V Hypervisor vendor: KVM Virtualization type: full L1d cache: 1.9 MiB (30 instances) L1i cache: 1.9 MiB (30 instances) L2 cache: 15 MiB (30 instances) L3 cache: 480 MiB (30 instances) NUMA node(s): 1 NUMA node0 CPU(s): 0-29 Vulnerability Gather data sampling: Not affected Vulnerability Indirect target selection: 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: Vulnerable: Safe RET, no microcode 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; IBRS_FW; STIBP disabled; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsa: Vulnerable: Clear CPU buffers attempted, no microcode Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Not affected

Versions of relevant libraries: [pip3] Could not collect [conda] Could not collect

cc @ezyang @gchanan @kadeng @msaroufim @csarofeen @ptrblck @eqy @nWEIdia @albanD @mruberry @jbschlosser @walterddr @mikaylagawarecki @xwang233