diff --git a/examples/configs/grpo_math_8B_megatron.yaml b/examples/configs/grpo_math_8B_megatron.yaml index fc839c8239..8273656ea7 100644 --- a/examples/configs/grpo_math_8B_megatron.yaml +++ b/examples/configs/grpo_math_8B_megatron.yaml @@ -54,7 +54,7 @@ policy: weight_decay_incr_style: "constant" lr_decay_style: "constant" lr_decay_iters: null - lr_warmup_iters: 50 + lr_warmup_iters: 13 lr_warmup_init: 3.0e-8 generation: