Segmentation fault in custom CUDA kernel .so during backprop

A native custom CUDA extension crashed the process during backward execution. The fault is inside the extension or its ABI boundary, not an NCCL collective simply because distributed ranks exit afterward. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent cuda-graphs failures.

The literal signature is "Segmentation fault in custom CUDA kernel .so during backprop". It is a environment failure associated with CUDA Graphs, custom kernels, and vision pipelines. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

A native custom CUDA extension crashed the process during backward execution. The fault is inside the extension or its ABI boundary, not an NCCL collective simply because distributed ranks exit afterward. The failure becomes visible at this call site because the operation first requires the missing resource, valid state, healthy peer, or correct result. Earlier log lines and a known-good control carry more causal value than the final wrapper exception.

• The workload stops or loses forward progress after emitting "Segmentation fault in custom CUDA kernel .so during backprop".
• A retry on the same configuration reproduces the failure because the causal state has not changed.
• The outer framework exception can hide the rank, node, allocation, or dependency that failed first.
• Increasing timeouts or reducing workload size can suppress the symptom without correcting the cause.

• CUDA Graphs, custom kernels, and vision pipelines
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

Apply the following checklist to a small reproduction: each box below is a positive signal that you are looking at this exact failure rather than a sibling in the same taxonomy.

• ✓Find the first occurrence of "Segmentation fault in custom CUDA kernel .so during backprop" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓rebuild the extension against the active PyTorch and CUDA ABI, validate tensor device, dtype, contiguity, shape, and lifetime, then run compute-sanitizer on the smallest failing case.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from last checkpoint before the failed backward step only after the control passes.

Segmentation fault in custom CUDA kernel .so during backprop

Timestamps and exact values vary across runs, but the pattern. An info-level start, an early WARN, an ERROR carrying the symptom. Is the actual fingerprint you should alert on. The Denpex platform flags this combination automatically.

reproduce on one GPU with CUDA_LAUNCH_BLOCKING=1 and enable Python faulthandler so the last extension call is preserved. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from last checkpoint before the failed backward step.

# Preserve evidence before restarting
rg -n -i 'error|exception|timeout|failed' <log-file>
nvidia-smi
python -m torch.utils.collect_env

# Find the exact signature in the complete log
rg -n -F -- "Segmentation fault in custom CUDA kernel .so during backprop" <log-file>

Adapt the snippet to your framework. The same pattern holds for PyTorch Lightning, Hugging Face Trainer, DeepSpeed, Megatron-LM, and vLLM training wrappers. Where the wrapper exposes a config flag (for examplelr_scheduler_type in Trainer), prefer the flag over the imperative API to keep the schedule declarative and reproducible.

Select the traceback or open the failed terminal, then run Denpex locally to see the initiating rank, collateral failures, exact fix, and verification command without uploading the log.

CUDA reports errors asynchronously, so the traceback usually points at whatever line synchronised next rather than the one at fault. The hub covers every common CUDA error and how to make it report honestly.

• A native custom CUDA extension crashed the process during backward execution. The fault is inside the extension or its ABI boundary, not an NCCL collective simply because distributed ranks exit afterward.
• The decisive evidence is the first log line that precedes "Segmentation fault in custom CUDA kernel .so during backprop" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.