Backprop through TransducerLoss creates NaN gradients

Start by running the provided Python reproduction in the NVIDIA PyTorch container and trace the TransducerLoss entry point during backpropagation. Done means the negative derivative case produces finite logits.grad values, while preserving the positive-derivative behavior; add a regression check if the project’s existing tests provide a suitable location.

Describe the Bug

I have a custom loss function my_custom_loss that uses TransducerLoss as one of its inputs. If the derivative of my_custom_loss with respect to the TransducerLoss is negative, then the gradients of my model are NaNs during backpropagation.

Minimal Steps/Code to Reproduce the Bug
Launch container via

docker run --gpus all -it --rm -v /local_dir:/container_dir nvcr.io/nvidia/pytorch:23.05-py3

where local_dir contains this python script:

#!/usr/bin/env python3
import torch
from apex.contrib.transducer import TransducerLoss

# Boilerplate to create input tensors with compatible dimensions:
logits = torch.rand(1, 10, 15, 20).cuda()
logits.requires_grad = True
logit_lens = torch.tensor([10], dtype=torch.int32).cuda()
y = torch.ones(1, 14, dtype=torch.int32).cuda()
y_lens = torch.tensor([14], dtype=torch.int32).cuda()
blank_idx = 19

t_loss = TransducerLoss()(logits, y, logit_lens, y_lens, blank_idx=blank_idx)
# In this example the derivative of my_custom_loss
# with respect to t_loss is -1:
my_custom_loss = 1000 - t_loss
# Will be a number somewhere near 950:
print(f"{my_custom_loss=}")
my_custom_loss.backward()
# Will be a tensor of NaNs:
print(f"{logits.grad=}")

In the container, run python /container_dir/name_of_script.py

Expected Behavior
logits.grad should be a tensor of numbers. Instead it is (incorrectly) a tensor of NaNs.

If instead my_custom_loss was defined as 1000 + t_loss, then logits.grad will correctly be a tensor of numbers when the script is run.

Environment

root@365e5bd35d3b:/workspace# python -m torch.utils.collect_env
Collecting environment information...
PyTorch version: 2.0.0
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.2 LTS (x86_64)
GCC version: (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0
Clang version: Could not collect
CMake version: version 3.24.1
Libc version: glibc-2.35

Python version: 3.10.6 (main, Mar 10 2023, 10:55:28) [GCC 11.3.0] (64-bit runtime)
Python platform: Linux-5.4.0-148-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA GeForce RTX 2080 Ti
Nvidia driver version: 530.30.02
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.1
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
Address sizes:                   39 bits physical, 48 bits virtual
Byte Order:                      Little Endian
CPU(s):                          12
On-line CPU(s) list:             0-11
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
CPU family:                      6
Model:                           158
Thread(s) per core:              2
Core(s) per socket:              6
Socket(s):                       1
Stepping:                        10
CPU max MHz:                     4700.0000
CPU min MHz:                     800.0000
BogoMIPS:                        7399.70
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 sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1d arch_capabilities
Virtualization:                  VT-x
L1d cache:                       192 KiB (6 instances)
L1i cache:                       192 KiB (6 instances)
L2 cache:                        1.5 MiB (6 instances)
L3 cache:                        12 MiB (1 instance)
NUMA node(s):                    1
NUMA node0 CPU(s):               0-11
Vulnerability Itlb multihit:     KVM: Vulnerable
Vulnerability L1tf:              Mitigation; PTE Inversion
Vulnerability Mds:               Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Meltdown:          Mitigation; PTI
Vulnerability Mmio stale data:   Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed:          Mitigation; IBRS
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; IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds:             Mitigation; Microcode
Vulnerability Tsx async abort:   Mitigation; Clear CPU buffers; SMT vulnerable

Versions of relevant libraries:
[pip3] numpy==1.22.2
[pip3] pytorch-quantization==2.1.2
[pip3] torch==2.0.0
[pip3] torch-tensorrt==1.4.0.dev0
[pip3] torchdata==0.6.0
[pip3] torchtext==0.15.1
[pip3] torchvision==0.15.1
[conda] Could not collect