1experiment:2 name: pretrain_enlarged102m3 seed: 14 model: configs/baselines/transformer_enlarged_102m.yaml5 data:6 train_bin: data/processed/fineweb_edu_300m/train.bin7 val_bin: data/processed/fineweb_edu_300m/val.bin8 train: &id0019 batch_size: 3210 seq_len: 51211 steps: 600012 lr: 0.000613 warmup: 30014 weight_decay: 0.115 grad_clip: 1.016 log_every: 2017 eval_every: 50018 eval_batches: 2019 ckpt_every: 100020model:21 name: transformer_enlarged_102m22 vocab_size: 3276823 d_model: 76824 n_layers: 1225 n_heads: 1226 n_kv_heads: 327 ffn_width: 214428 max_seq_len: 409629 rope_theta: 10000.030 norm_eps: 1.0e-0531 tie_embeddings: true32 init_std: 0.0233 use_workspace: false34 w: 25635 ffn_w: 76836 n_slots: 1637 n_editor_blocks: 238 n_editor_heads: 439 r_max: 840 eta_max: 0.2541 clip_c: 1.042 clip_eps: 1.0e-0643 readout_zero_init: true44 readout_init_std: 0.0245 readout_scale: 1.046 controller_hidden: 047 ctrl_input: z48 ctrl_use_entropy: false49 lambda_cost: 0.050 kappa_risk: 0.051 dense_writes: false52resolved_train: *id00153