Tag: LLM training

Monitoring Loss and Perplexity: A Practical Guide to LLM Training Signals

Learn how to interpret cross-entropy loss and perplexity during LLM training. Discover practical tips for reading training signals, avoiding overfitting, and diagnosing model health effectively.

Read more

Checkpoint Averaging and EMA: Stabilizing Large Language Model Training

Checkpoint averaging and EMA stabilize large language model training by combining model snapshots to improve performance and reduce variance - delivering 1-2% gains with minimal overhead. Now standard for models over 1B parameters.

Read more