Results for "compute-data-performance"

Empirical laws linking model size, data, compute to performance.

Increasing model capacity via compute.

Regulating access to large-scale compute.

Letting an LLM call external functions/APIs to fetch data, compute, or take actions, improving reliability.

Scaling law optimizing compute vs data.

When information from evaluation data improperly influences training, inflating reported performance.

Maliciously inserting or altering training data to implant backdoors or degrade performance.

Increasing performance via more data.

Learning a function from input-output pairs (labeled data), optimizing performance on predicting outputs for unseen inputs.

A mismatch between training and deployment data distributions that can degrade model performance.

Updating a pretrained model’s weights on task-specific data to improve performance or adapt style/behavior.

Standardized documentation describing intended use, performance, limitations, data, and ethical considerations.

Systematic review of model/data processes to ensure performance, fairness, security, and policy compliance.

Number of samples per gradient update; impacts compute efficiency, generalization, and stability.

Techniques that fine-tune small additional components rather than all weights to reduce compute and storage.

Storing results to reduce compute.

Stored compute or algorithms enabling rapid jumps.

A robust evaluation technique that trains/evaluates across multiple splits to estimate performance variability.

Often more informative than ROC on imbalanced datasets; focuses on positive class performance.

Halting training when validation performance stops improving to reduce overfitting.

Achieving task performance by providing a small number of examples inside the prompt without weight updates.

Training a smaller “student” model to mimic a larger “teacher,” often improving efficiency while retaining performance.

Performance drop when moving from simulation to reality.

Unequal performance across demographic groups.

Tradeoff between safety and performance.

Learning structure from unlabeled data, such as discovering groups, compressing representations, or modeling data distributions.

Minimizing average loss on training data; can overfit when data is limited or biased.

When a model fits noise/idiosyncrasies of training data and performs poorly on unseen data.

Training across many devices/silos without centralizing raw data; aggregates updates, not data.

Protecting data during network transfer and while stored; essential for ML pipelines handling sensitive data.