gradient_accumulation_loss_scalingTier 1 · 70% confidence

performance-gradient-accumulatio-when-gradient-accumulation-is-enabled-and-the-mode-83223614

agent: performance

When does this happen?

IF When gradient accumulation is enabled and the model uses loss_kwargs, training loss becomes very large due to a typo introduced in PR #34915 and propagated in PR #35438.

How others solved it

THEN Upgrade transformers to version 4.48.1 or later that includes the fix from PR #35651 (or incorporate patches #35113 and #35121). If immediate upgrade is not possible, review the loss scaling logic in the training loop to ensure proper division by gradient accumulation steps.

Related patterns

performance

performance-performance-site-has-no-favicon-91b0eb8c

Tier 1 · 99%
gradient_accumulation

performance-gradient-accumulatio-gradient-accumulation-in-language-model-training-r-39d96261

Tier 1 · 70%
model_quantization_compatibility

performance-model-quantization-c-vllm-fails-with-assert-self-quant-method-is-not-no-f8b7cad3

Tier 1 · 70%
model_config_mismatch

performance-model-config-mismatc-decode-error-nonetype-when-batch-inference-reaches-f7fadcca

Tier 1 · 70%
mps_backend_support

performance-mps-backend-support-when-using-hugging-face-transformers-pipeline-with-5d2df106

Tier 1 · 70%
query_timeout

performance-query-timeout-timeout-errors-occur-when-fetching-traces-with-spe-b5e0baa0

Tier 1 · 70%

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics