List your company

Language Modeling on LLM (val)

1.3364LossdynActGLU(Swish)
1.3295681.3756841.42181.467916Mar 23, 2026

Evaluation results

MethodLinks
dynActGLU(Swish)Activation=dynActGLU(Swish), Training steps=343002026.031.33643.8056
Baseline SwiGLUActivation=SwiGLU, Training steps=343002026.031.33743.8091
dynActGLU(Swish)Activation=dynActGLU(Swish), Training steps=56202026.031.39794.0467
Baseline SwiGLUActivation=Baseline SwiGLU, Training steps=56202026.031.50724.5142