Rotary embedding dimension must be divisible by head dimension

The rotary embedding dimension does not divide evenly into the attention head dimension. This is a configuration arithmetic error, caught at construction: hidden_size / num_attention_heads must be compatible with the rotary percentage. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent megatron failures.

The literal signature is "Rotary embedding dimension must be divisible by head dimension". It is a distributed training failure associated with Megatron Core tensor and pipeline parallelism. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

The rotary embedding dimension does not divide evenly into the attention head dimension. This is a configuration arithmetic error, caught at construction: hidden_size / num_attention_heads must be compatible with the rotary percentage. 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 "Rotary embedding dimension must be divisible by head dimension".
• 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.

• Megatron Core tensor and pipeline parallelism
• 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 "Rotary embedding dimension must be divisible by head dimension" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓the usual cause is changing num_attention_heads or hidden_size without re-deriving the rest, or importing a config from a model with a different head layout. Print all four values together, the mismatch is obvious once they are side by side.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from job start after correcting the config only after the control passes.

Rotary embedding dimension must be divisible by head dimension

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.

check the arithmetic. head_dim = hidden_size / num_attention_heads, and rotary_dim = head_dim * rotary_percent must be an even integer. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from job start after correcting the config.

# 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 -- "Rotary embedding dimension must be divisible by head dimension" <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.

• The rotary embedding dimension does not divide evenly into the attention head dimension. This is a configuration arithmetic error, caught at construction: hidden_size / num_attention_heads must be compatible with the rotary percentage.
• The decisive evidence is the first log line that precedes "Rotary embedding dimension must be divisible by head dimension" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.