NaN Loss During Training
NaN loss corrupts training state. Denpex traces NaN propagation to the originating layer.
NaN loss corrupts training state.
What this failure is
NaN Loss During Training is a Training Stability failure seen during ML training runs. NaN loss corrupts training state. Denpex traces NaN propagation to the originating layer. Common tags: Nan, Loss, Numerical Stability, Precision.
Is this what broke your run? Paste your log.
You're reading about NaN Loss During Training. Paste your own crash log or traceback below and get the real root cause for YOUR run, not this generic entry. No account, no card. Logs are masked at ingress and never saved to account history.
Want 14 days on the Scale plan?
Request an evaluation code. A verified workplace organization activates up to 50 diagnoses a day, alerts, history, and follow-up questions. No credit card or automatic subscription.
Why it happens (the mechanism)
Numerical underflow in softmax. Division by zero in attention. Log of zero in log-probability. AMP gradient scaler overflow. Taken together, these mechanisms explain why the failure is reproducible, why it tends to surface on specific workloads or scales, and why generic mitigation attempts often fall short without addressing the underlying cause.
What you'll observe
- Loss becomes NaN after training steps
- All subsequent gradients become NaN
- Run is unrecoverable without checkpoint restore
Common symptoms and what they mean
| Symptom | Why it happens |
|---|---|
| loss: nan in training logs | Numerical underflow in softmax |
| torch.autograd detects NaN | Division by zero in attention |
| Model outputs become NaN | Log of zero in log-probability |
Which systems are affected
- Large language models
- Mixed-precision training
- Models with softmax or log_softmax
How to confirm this is the problem
Use this checklist to test the hypothesis against a small reproduction. No single line proves the root cause, so preserve the preceding events and compare one variable at a time.
- ✓Reproduce the failure from a clean checkpoint/seed: the symptom must appear without warm-up state from a previous run.
- ✓Verified signal present: loss: nan in training logs
- ✓Verified signal present: torch.autograd detects NaN
- ✓Verified signal present: Model outputs become NaN
- ✓A targeted fix from the "How to fix it" section eliminates or substantially reduces the symptom within one validation pass.
The fix and the prevention pattern
The root cause is on this page and stays free. A free account adds the exact remediation steps, saved history, and the fix on every entry in the encyclopedia.
Sign up free. Unlock the full analysisNo credit card. Daily allowance follows verified trust tier. Instant access.
Diagnose this failure in VS Code
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.
Install the free VS Code extensionRelated failures to investigate next
Root cause
- Numerical underflow in softmax
- Division by zero in attention
- Log of zero in log-probability
- AMP gradient scaler overflow
The fix and how to prevent it
Evaluate Denpex on your own logs
Request a Scale evaluation code. A verified workplace organization activates 14 days with up to 50 diagnoses a day. Every account keeps its current diagnosis allowance and gets a verification path. No card or automatic subscription.
Don't just read the fix, diagnose your run
The encyclopedia tells you what went wrong. Denpex tells you what went wrong in YOUR training run. With your logs, your config, and your stack.