SPB Git

spb/wp31_uqo

Public
1experiment:2  name: p4_editor_v1_Ustd023  seed: 14  mode: prisme5  model:6    name: tiny_p47    vocab_size: 1288    d_model: 1929    n_layers: 410    n_heads: 611    n_kv_heads: 212    ffn_width: 51213    max_seq_len: 32014    w: 6415    ffn_w: 19216    n_slots: 1617    n_editor_heads: 218    r_max: 819  init:20    backbone_checkpoint: experiments/runs/p4_tiny_backbone_seed01_20261029232626/checkpoint21    freeze_backbone: true22    auto_readout_scale: true23  data:24    kind: synthetic25    tokenizer: char26    splits_dir: data/splits/synthetic_v127    loss_span: answer28    eval_splits:29    - dev_iid30    - dev_template31    - dev_depth32    - dev_compose33  budgets:34  - 035  - 136  - 237  - 438  - 839  budget_probs:40  - 0.241  - 0.242  - 0.243  - 0.244  - 0.245  budget_per: position46  train:47    batch_size: 3248    seq_len: 32049    steps: 200050    lr: 0.00151    warmup: 10052    weight_decay: 0.153    grad_clip: 1.054    log_every: 5055    eval_every: 50056    eval_batches: 857    ckpt_every: 150058  em_budgets:59  - 060  - 261  - 862  em_max_items: 40063  model_overrides:64    readout_zero_init: false65    readout_init_std: 0.266model:67  name: tiny_p468  vocab_size: 12869  d_model: 19270  n_layers: 471  n_heads: 672  n_kv_heads: 273  ffn_width: 51274  max_seq_len: 32075  rope_theta: 10000.076  norm_eps: 1.0e-0577  tie_embeddings: true78  init_std: 0.0279  use_workspace: true80  w: 6481  ffn_w: 19282  n_slots: 1683  n_editor_blocks: 284  n_editor_heads: 285  r_max: 886  eta_max: 0.2587  clip_c: 1.088  clip_eps: 1.0e-0689  readout_zero_init: false90  readout_init_std: 0.291  readout_scale: 1.092  controller_hidden: 093  lambda_cost: 0.094  kappa_risk: 0.095  dense_writes: false96resolved_train:97  batch_size: 3298  seq_len: 32099  steps: 2000100  lr: 0.001101  warmup: 100102  weight_decay: 0.1103  grad_clip: 1.0104  log_every: 50105  eval_every: 500106  eval_batches: 8107  ckpt_every: 1500108
Files · main
Menu

Theme