DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter

ZeRO-2 batches gradients into a fixed communication bucket before reduce-scatter. A gradient tensor larger than the bucket, or too many arriving at once, exhausts it and the reduction cannot proceed. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent deepspeed failures.

Symptom

DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter

Root cause

ZeRO-2 batches gradients into a fixed communication bucket before reduce-scatter. A gradient tensor larger than the bucket, or too many arriving at once, exhausts it and the reduction cannot proceed. The decisive evidence is the first log line that precedes "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter" and differs from a healthy run.

Recommended fix

raise allreduce_bucket_size (and reduce_bucket_size) in the DeepSpeed config. 2e8 is a common working value; the default is often too small for large embeddings.

How Denpex helps

Denpex investigates DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter using the evidence you provide or your connected workload collects. Earlier rank, host or application evidence is needed to distinguish an initiating failure from a downstream report.

The literal signature is "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter". It is a distributed training failure associated with DeepSpeed ZeRO and optimizer sharding. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

ZeRO-2 batches gradients into a fixed communication bucket before reduce-scatter. A gradient tensor larger than the bucket, or too many arriving at once, exhausts it and the reduction cannot proceed. 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 "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter".
• 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.

• DeepSpeed ZeRO and optimizer sharding
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

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.

• ✓Find the first occurrence of "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓find the largest single gradient, max(p.numel() for p in model.parameters()). If one tensor exceeds the bucket, no amount of tuning batch count helps; the bucket must exceed the largest tensor.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from latest checkpoint only after the control passes.

• ZeRO-2 batches gradients into a fixed communication bucket before reduce-scatter. A gradient tensor larger than the bucket, or too many arriving at once, exhausts it and the reduction cannot proceed.
• The decisive evidence is the first log line that precedes "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.

DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter

Use this text as a lookup key in logs and upstream issue trackers. It is not presented as a captured customer log. Confirm the cause from your own preceding events, versions, configuration and the cited references.

raise allreduce_bucket_size (and reduce_bucket_size) in the DeepSpeed config. 2e8 is a common working value; the default is often too small for large embeddings. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from latest checkpoint.

# 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 -- "DeepSpeed ZeRO-2 tensor bucket buffer exhausted during reduce_scatter" <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.

DeepSpeed changes when parameters, gradients and optimizer state are created, partitioned, gathered and offloaded. The hub separates ZeRO, memory, checkpoint and pipeline failures by lifecycle phase.