Unsloth Fused Loss Breaks with Transformers average_tokens_across_devices=True
Multi-GPU fine-tuning with Unsloth breaks when the Transformers default average_tokens_across_devices=True multiplies the loss by num_processes. The Unsloth fused-loss backward does not expect that scaling and the loss tensor is corrupted (becomes an int / loses grad). Setting average_tokens_across_devices=False fixes it.
Multi-GPU fine-tuning with Unsloth breaks when the Transformers default average_tokens_across_devices=True multiplies the loss by num_processes.
What this failure is
Unsloth Fused Loss Breaks with Transformers average_tokens_across_devices=True is a Training Stability failure seen during ML training runs. Multi-GPU fine-tuning with Unsloth breaks when the Transformers default average_tokens_across_devices=True multiplies the loss by num_processes. The Unsloth fused-loss backward does not expect that scaling and the loss tensor is corrupted (becomes an int / loses grad). Setting average_tokens_across_devices=False fixes it. Common tags: Unsloth, Transformers, Average_tokens_across_devices, Fused Loss.
Is this what broke your run? Paste your log.
You're reading about Unsloth Fused Loss Breaks with Transformers average_tokens_across_devices=True. 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.
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.
Why it happens (the mechanism)
Newer Transformers scales the loss by num_processes to average tokens across devices. Unsloth's fused-loss path multiplies/handles the loss assuming the unscaled value, so the extra factor corrupts the tensor type and gradient. 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
- UnslothFusedLossBackward errors or the loss becomes an int on multi-GPU
- Loss is the wrong scale across devices
- Backward fails on the fused loss tensor
Common symptoms and what they mean
| Symptom | Why it happens |
|---|---|
| Loss multiplied by a factor (num_processes / n_gpu) before backward | Newer Transformers scales the loss by num_processes to average tokens across devices |
| Loss turns into an int or loses requires_grad | Unsloth's fused-loss path multiplies/handles the loss assuming the unscaled value, so the extra factor corrupts the tensor type and gradient |
| Only on multi-GPU / distributed runs | Newer Transformers scales the loss by num_processes to average tokens across devices |
Which systems are affected
- Unsloth fused cross-entropy on multi-GPU
- Recent Transformers with average_tokens_across_devices defaulting True
- Distributed SFT/GRPO with Unsloth
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: Loss multiplied by a factor (num_processes / n_gpu) before backward
- ✓Verified signal present: Loss turns into an int or loses requires_grad
- ✓Verified signal present: Only on multi-GPU / distributed runs
- ✓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 analysisNo 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 extensionRelated failures to investigate next
Root cause
- Newer Transformers scales the loss by num_processes to average tokens across devices
- Unsloth's fused-loss path multiplies/handles the loss assuming the unscaled value, so the extra factor corrupts the tensor type and gradient
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.
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.