Skip to content

DeepSpeed ZeRO-3 High GPU Memory / OOM Loading a Large Model Without zero.Init

Fine-tuning a large model (e.g. Flan-T5-XXL 11B) with ZeRO-3 uses far more GPU memory than expected or OOMs at load because the model was constructed outside DeepSpeed's zero.Init context. Every rank materializes the full dense model before partitioning. Build the model under zero.Init (or keep HfDeepSpeedConfig alive before from_pretrained).

Quick answer

Fine-tuning a large model (e.

Memory#deepspeed#zero-3#zero-init#hf-deepspeed-config#oom#memory

What this failure is

DeepSpeed ZeRO-3 High GPU Memory / OOM Loading a Large Model Without zero.Init is a Memory failure seen during ML training runs. Fine-tuning a large model (e.g. Flan-T5-XXL 11B) with ZeRO-3 uses far more GPU memory than expected or OOMs at load because the model was constructed outside DeepSpeed's zero.Init context. Every rank materializes the full dense model before partitioning. Build the model under zero.Init (or keep HfDeepSpeedConfig alive before from_pretrained). Common tags: Deepspeed, Zero 3, Zero Init, Hf Deepspeed Config.

Live diagnosis, no signup

Is this what broke your run? Paste your log.

You're reading about DeepSpeed ZeRO-3 High GPU Memory / OOM Loading a Large Model Without zero.Init. 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)

The model was instantiated outside deepspeed.zero.Init(), so each rank loaded the complete dense parameters into GPU memory before ZeRO-3 could shard them. Without zero.Init the partitioning happens too late to avoid the full-model memory spike. 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

  • ZeRO-3 uses much more per-GPU memory than the partitioned size implies
  • OOM while loading the model with from_pretrained
  • Memory does not drop after ZeRO-3 'partitions' the model

Common symptoms and what they mean

SymptomWhy it happens
CUDA OOM during AutoModel.from_pretrained under ZeRO-3The model was instantiated outside deepspeed.zero.Init(), so each rank loaded the complete dense parameters into GPU memory before ZeRO-3 could shard them
Full dense model materialized on every GPU before partitioningWithout zero.Init the partitioning happens too late to avoid the full-model memory spike
11B+ model won't load on hardware that should fit the partitioned modelThe model was instantiated outside deepspeed.zero.Init(), so each rank loaded the complete dense parameters into GPU memory before ZeRO-3 could shard them

Which systems are affected

  • DeepSpeed ZeRO-3 fine-tuning of large models (Flan-T5-XXL, etc.)
  • HuggingFace Trainer + DeepSpeed where zero.Init isn't enabled
  • Param/optimizer CPU offload setups

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: CUDA OOM during AutoModel.from_pretrained under ZeRO-3
  • Verified signal present: Full dense model materialized on every GPU before partitioning
  • Verified signal present: 11B+ model won't load on hardware that should fit the partitioned model
  • 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

DeepSpeed errors in context

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.

Compare every deepspeed error side by side

Root cause

  • The model was instantiated outside deepspeed.zero.Init(), so each rank loaded the complete dense parameters into GPU memory before ZeRO-3 could shard them
  • Without zero.Init the partitioning happens too late to avoid the full-model memory spike

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.