Skip to content

Checkpoint Corruption

Corrupted checkpoints silently poison resumed training with bad weights. Denpex validates checkpoint integrity before resume.

Quick answer

Corrupted checkpoints silently poison resumed training with bad weights.

Data Integrity#checkpoint#corruption#data-integrity#save#resume#storage

What this failure is

Checkpoint Corruption is a Data Integrity failure seen during ML training runs. Corrupted checkpoints silently poison resumed training with bad weights. Denpex validates checkpoint integrity before resume. Common tags: Checkpoint, Corruption, Data Integrity, Save.

Live diagnosis, no signup

Is this what broke your run? Paste your log.

You're reading about Checkpoint Corruption. 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)

Torn writes: the checkpoint save process is interrupted (node failure, OOM killer, NCCL timeout) mid-write, leaving an incomplete file that passes filesystem validation. Shard inconsistency: in distributed checkpointing, one rank's shard save succeeds while another's fails, leaving an incomplete sharded checkpoint. Bit rot / silent data corruption: storage media flips bits in the serialized tensor data without filesystem error detection. World size mismatch: checkpoint saved with N GPUs is loaded with M GPUs, causing optimizer state partition reassignment errors. 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

  • Resumed training produces worse metrics than original
  • Checkpoint load succeeds but model output is garbage
  • Checkpoint file size differs between identical saves

Common symptoms and what they mean

SymptomWhy it happens
Loading checkpoint succeeds but model.eval() produces constant outputTorn writes: the checkpoint save process is interrupted (node failure, OOM killer, NCCL timeout) mid-write, leaving an incomplete file that passes filesystem validation
RuntimeError: 'lengths' argument in 'EmbeddingBag' is empty after checkpoint loadShard inconsistency: in distributed checkpointing, one rank's shard save succeeds while another's fails, leaving an incomplete sharded checkpoint
Optimizer state dictionary shape mismatch on resume with different world sizeBit rot / silent data corruption: storage media flips bits in the serialized tensor data without filesystem error detection
Checkpoint file hash differs after repeated saves of identical model stateWorld size mismatch: checkpoint saved with N GPUs is loaded with M GPUs, causing optimizer state partition reassignment errors

Which systems are affected

  • Distributed checkpointing with torch.distributed.checkpoint
  • Hugging Face Trainer saving with save_safetensors=False
  • NFS-mounted checkpoint directories with concurrent writers
  • Cloud storage backed checkpoints (S3, GCS, Azure Blob) with partial uploads

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: Loading checkpoint succeeds but model.eval() produces constant output
  • Verified signal present: RuntimeError: 'lengths' argument in 'EmbeddingBag' is empty after checkpoint load
  • Verified signal present: Optimizer state dictionary shape mismatch on resume with different world size
  • Verified signal present: Checkpoint file hash differs after repeated saves of identical model state
  • 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

Root cause

  • Torn writes: the checkpoint save process is interrupted (node failure, OOM killer, NCCL timeout) mid-write, leaving an incomplete file that passes filesystem validation
  • Shard inconsistency: in distributed checkpointing, one rank's shard save succeeds while another's fails, leaving an incomplete sharded checkpoint
  • Bit rot / silent data corruption: storage media flips bits in the serialized tensor data without filesystem error detection
  • World size mismatch: checkpoint saved with N GPUs is loaded with M GPUs, causing optimizer state partition reassignment errors

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.