A Quick Diagnostic Guide to 4 Common ML Training Errors and Their Causes
cneuralnetwork · x · 2026-11-12
A handy cheat sheet maps the 4 most common ML training errors to their usual causes:
- CUDA OOM: batch too large, model too big for the GPU, or unreleased tensors/activations leaking memory.
- Loss → NaN: learning rate too high, numerical instability (e.g., log(0), division by zero), exploding gradients.
- Shape mismatch: wrong tensor dimensions (e.g., [B,C,H,W] vs [B,H,W,C]), batch/sequence length misalignment, missing reshape/unsqueeze/transpose.
- Accuracy stuck: underfitting (too simple or over-regularized), bugs in data/labels/metric computation, wrong learning rate or training setup.
More from Research
- Token pooling combines well with embedding quantization and Matryoshka — antoine_chaffin · 2026-11-12
- FAME: exploration planning method transfers across scenes, beating NBP by 6.6 points on reconstruction — jiqizhixin · 2026-11-12
- E11 Bio hiring computational scientist for petabyte imaging and agentic AI labs — anderssandberg · 2026-11-12
- EPFL: Claude writes EEG network weights by reasoning, matching gradient training with 6x fewer subjects — epfl-ch · 2026-11-12
- Differentiable SSSS framework brings single-image real-time skin digitization to game engines — ssh4net · 2026-11-12
- NeurIPS paper proves when RNN+MLP can provably recover true robot state — chriswolfvision · 2026-11-12