diff --git a/llmc/schedulers.h b/llmc/schedulers.h index bf469aa..bd0f5f4 100644 --- a/llmc/schedulers.h +++ b/llmc/schedulers.h @@ -1,7 +1,6 @@ - - -// Cosine learning rate scheduler - +/* +Implements various learning rate schedulers. +*/ #ifndef SCHEDULERS_H #define SCHEDULERS_H @@ -16,7 +15,6 @@ typedef struct { float final_learning_rate_frac; } CosineLearningRateScheduler; - // learning rate schedule: warmup linearly to max LR, then cosine decay to LR * final_learning_rate_frac float get_learning_rate(CosineLearningRateScheduler *scheduler, int step) { float step_learning_rate = scheduler->learning_rate;