====================weight_decay_scale====================
decoder/emb/token_emb/weight: 1
decoder/output_norm/scale: 1
decoder/transformer/repeat/layer/feed_forward/linear1_0/weight: 1
decoder/transformer/repeat/layer/feed_forward/linear1_1/weight: 1
decoder/transformer/repeat/layer/feed_forward/linear2/weight: 1
decoder/transformer/repeat/layer/feed_forward/norm/scale: 1
decoder/transformer/repeat/layer/self_attention/attention/i_proj/i_proj/qkv_proj/weight: 1
decoder/transformer/repeat/layer/self_attention/attention/o_proj/weight: 1
decoder/transformer/repeat/layer/self_attention/norm/scale: 1
