What Is AI Model Pruning?
Model pruning removes or suppresses selected weights, connections, channels, or other structures to make a model smaller or cheaper.
Don't Just Read it – Explore It
Model pruning removes or suppresses selected weights, connections, channels, or other structures to make a model smaller or cheaper.
MLOps applies engineering practices to the full machine-learning lifecycle, including data, training, deployment, monitoring, and retraining.
Feature engineering turns raw observations into model inputs that expose useful structure while preserving production-time validity.
A feature store manages reusable feature definitions and values so training and online prediction use consistent inputs.
Data leakage occurs when model training or evaluation uses information that would not be legitimately available at prediction time.
Class imbalance occurs when some target classes have far fewer examples than others, making aggregate accuracy potentially misleading.
Data labeling assigns target values or annotations to examples so supervised systems can learn and be evaluated.
BLEU is an automatic text-generation metric that compares n-gram overlap between candidate translations and reference translations.
ROUGE is a family of metrics that compares overlap between generated text and reference text, especially for summarization.
Perplexity summarizes how much probability a language model assigns to a sequence, with lower values indicating better prediction on that data.
Sampling temperature rescales token logits before sampling, changing how concentrated or spread out the next-token probabilities are.
Top-k and top-p sampling restrict the tokens eligible for random selection before the model chooses its next token.