SPB Git

spb/wp31_uqo

Public
1experiment:2  name: labels_mini3  seed: 14  mode: prisme5  model: from_checkpoint6  checkpoint: experiments/runs/mini_editor_seed01_20261030005118/checkpoint7  use_policy: false8  data:9    kind: synthetic10    tokenizer: char11    splits_dir: data/splits/synthetic_v112    loss_span: answer13  collect:14    batch_size: 1615    seq_len: 32016    max_prefix: 217    audit_every: 218  partitions:19    train:20      split: ctrl_train21      n_batches: 1222      stream_seed: 1123    calib:24      split: calib25      n_batches: 426      stream_seed: 1227model:28  name: mini29  vocab_size: 12830  d_model: 9631  n_layers: 232  n_heads: 433  n_kv_heads: 234  ffn_width: 25635  max_seq_len: 32036  rope_theta: 10000.037  norm_eps: 1.0e-0538  tie_embeddings: true39  init_std: 0.0240  use_workspace: true41  w: 3242  ffn_w: 9643  n_slots: 1644  n_editor_blocks: 245  n_editor_heads: 246  r_max: 847  eta_max: 0.2548  clip_c: 1.049  clip_eps: 1.0e-0650  readout_zero_init: false51  readout_init_std: 0.252  readout_scale: 0.1196771636605262853  controller_hidden: 054  ctrl_input: h55  ctrl_use_entropy: true56  lambda_cost: 0.057  kappa_risk: 0.058  dense_writes: false59
Files · main
Menu

Theme