SPB Git

spb/wp31_uqo

Public
693 B · 44 lines yaml
1name: mini_backbone2seed: 13mode: backbone_only4model:5  name: mini6  vocab_size: 1287  d_model: 968  n_layers: 29  n_heads: 410  n_kv_heads: 211  ffn_width: 25612  max_seq_len: 32013  w: 3214  ffn_w: 9615  n_slots: 1616  n_editor_heads: 217  r_max: 818  readout_zero_init: false19  readout_init_std: 0.220  ctrl_input: h21  ctrl_use_entropy: true22init:23  freeze_backbone: false24data:25  kind: synthetic26  tokenizer: char27  splits_dir: data/splits/synthetic_v128  loss_span: all29  eval_splits:30  - dev_iid31train:32  batch_size: 1633  seq_len: 32034  steps: 15035  lr: 0.00136  warmup: 1037  weight_decay: 0.138  grad_clip: 1.039  log_every: 5040  eval_every: 15041  eval_batches: 242  ckpt_every: 15043em_max_items: 3244
Files · a2c46f03643fdd75e8e87595931d35e18c590d57
Menu

Theme