List your company

Training Stability Analysis on Llama-2 7B pre-training

0Number of SpikesAdaGC
-0.362.074.56.93Feb 16, 2025

Evaluation results

MethodLinks
AdaGCTotal Steps=9K2025.0200
AdaGCTotal Steps=9K2025.0200
WeSaR-GlobalGCTotal Steps=9K2025.0211.11
SPAMTotal Steps=9K2025.0233.33
GlobalGCTotal Steps=9K2025.0233.33
ClippyTotal Steps=9K2025.0233.33
ScaledEmbed-GlobalGCTotal Steps=9K2025.0288.89
AGCTotal Steps=9K2025.0288.89
ClipByValueTotal Steps=9K2025.02910