Skip to content

Gradient Explosion Caused by Corrupted All-Reduce (Not Learning Rate)

Gradient norms explode suddenly while loss remains normal because the cross-rank reduction itself is corrupting values. In-network reduction (SHARP) faults, NCCL data corruption over a marginal link, or a mixed-precision comm dtype bug. Per-rank local gradients are healthy; only the aggregated result is wrong.

Quick answer

Gradient norms explode suddenly while loss remains normal because the cross-rank reduction itself is corrupting values. In-network reduction (SHARP) faults, NCCL data corruption over a marginal link, or a mixed-precision comm dtype bug.

Training Stability#gradient-explosion#nccl#sharp#infiniband#allreduce#grad-norm

What this failure is

Gradient Explosion Caused by Corrupted All-Reduce (Not Learning Rate) is a Training Stability failure seen during ML training runs. Gradient norms explode suddenly while loss remains normal because the cross-rank reduction itself is corrupting values. In-network reduction (SHARP) faults, NCCL data corruption over a marginal link, or a mixed-precision comm dtype bug. Per-rank local gradients are healthy; only the aggregated result is wrong. Common tags: Gradient Explosion, Nccl, Sharp, Infiniband.

Live diagnosis, no signup

Is this what broke your run? Paste your log.

You're reading about Gradient Explosion Caused by Corrupted All-Reduce (Not Learning Rate). 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.

training_logs.txt
No log to hand? Try one:

3 free diagnoses/day

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.

Evaluate one incident

Why it happens (the mechanism)

Switch ALU corruption when NCCL_SHARP is enabled on faulty firmware. Bit corruption over a link failing CRC-silently. Comm-hook dtype truncation overflowing fp16 during reduce. 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

  • grad_norm jumps 3-6 orders of magnitude in one step with no LR change
  • loss at the same step is still normal (divergence follows a step later)
  • per-rank pre-reduce norms differ from post-reduce values

Common symptoms and what they mean

SymptomWhy it happens
grad_norm=41211.7 after steps of ~1-2 with unchanged scheduleSwitch ALU corruption when NCCL_SHARP is enabled on faulty firmware
allreduce result mismatch when verified rank-by-rankBit corruption over a link failing CRC-silently
occurs only at specific scale/topology (SHARP-enabled fabrics)comm-hook dtype truncation overflowing fp16 during reduce

Which systems are affected

  • NCCL with SHARP/in-network computing
  • InfiniBand fabrics with switch-resident reduction
  • mixed bf16/fp32 comm hooks

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: grad_norm=41211.7 after steps of ~1-2 with unchanged schedule
  • Verified signal present: allreduce result mismatch when verified rank-by-rank
  • Verified signal present: occurs only at specific scale/topology (SHARP-enabled fabrics)
  • 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 analysis

No 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 extension

NCCL errors in context

NCCL is where a distributed job reports failure, which is not the same as where it failed. The hub lists every common NCCL error next to what it actually indicates, and the environment variables that tell them apart.

Compare every nccl error side by side

Root cause

  • Switch ALU corruption when NCCL_SHARP is enabled on faulty firmware
  • Bit corruption over a link failing CRC-silently
  • comm-hook dtype truncation overflowing fp16 during reduce

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.

We send a single-use code tied to that address. Static provider and TLD rules do not reject valid addresses. Account trust determines the benefit after signup.

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.