1experiment:2 name: p4_tiny_backbone_v23 seed: 14 mode: backbone_only5 model:6 name: tiny_p4v27 vocab_size: 1288 d_model: 2569 n_layers: 610 n_heads: 811 n_kv_heads: 412 ffn_width: 76813 max_seq_len: 32014 w: 6415 ffn_w: 19216 n_slots: 1617 n_editor_heads: 218 r_max: 819 init:20 freeze_backbone: false21 data:22 kind: synthetic23 tokenizer: char24 splits_dir: data/splits/synthetic_v125 loss_span: all26 eval_splits:27 - dev_iid28 - dev_template29 - dev_depth30 - dev_compose31 train:32 batch_size: 6433 seq_len: 32034 steps: 1500035 lr: 0.00136 warmup: 30037 weight_decay: 0.138 grad_clip: 1.039 log_every: 10040 eval_every: 100041 eval_batches: 1042 ckpt_every: 500043 em_max_items: 80044model:45 name: tiny_p4v246 vocab_size: 12847 d_model: 25648 n_layers: 649 n_heads: 850 n_kv_heads: 451 ffn_width: 76852 max_seq_len: 32053 rope_theta: 10000.054 norm_eps: 1.0e-0555 tie_embeddings: true56 init_std: 0.0257 use_workspace: false58 w: 6459 ffn_w: 19260 n_slots: 1661 n_editor_blocks: 262 n_editor_heads: 263 r_max: 864 eta_max: 0.2565 clip_c: 1.066 clip_eps: 1.0e-0667 readout_zero_init: true68 controller_hidden: 069 lambda_cost: 0.070 kappa_risk: 0.071 dense_writes: false72resolved_train:73 batch_size: 6474 seq_len: 32075 steps: 1500076 lr: 0.00177 warmup: 30078 weight_decay: 0.179 grad_clip: 1.080 log_every: 10081 eval_every: 100082 eval_batches: 1083 ckpt_every: 500084