SPB Git

spb/wp31_uqo

Public
914 B · 62 lines yaml
1name: mini_editor2seed: 13mode: prisme4model: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  backbone_checkpoint: experiments/runs/mini_backbone_seed01_20261030005114/checkpoint24  freeze_backbone: true25  auto_readout_scale: true26data:27  kind: synthetic28  tokenizer: char29  splits_dir: data/splits/synthetic_v130  loss_span: answer31  eval_splits:32  - dev_iid33train:34  batch_size: 1635  seq_len: 32036  steps: 12037  lr: 0.00138  warmup: 1039  weight_decay: 0.140  grad_clip: 1.041  log_every: 5042  eval_every: 12043  eval_batches: 244  ckpt_every: 12045em_max_items: 3246budgets:47- 048- 149- 250- 451- 852budget_probs:53- 0.254- 0.255- 0.256- 0.257- 0.258budget_per: position59em_budgets:60- 061- 262
Files · 20e88d79efe28fb13610284e7be1ef0ef40c265d
Menu

Theme