SPB Git

spb/wp31_uqo

Public
1experiment:2  name: controller_103m3  seed: 14  mode: prisme5  model: from_checkpoint6  checkpoint: experiments/runs/editor_pilot_103m_lr3e4_seed01_20261030021200/checkpoint7  labels:8    train: experiments/runs/labels_103m_seed01_20261030034314/labels_train.pt9    calib: experiments/runs/labels_103m_seed01_20261030034314/labels_calib.pt10  train:11    steps: 150012    lr: 0.00113    weight_decay: 0.0114    alpha: 1.015    beta: 0.516    gamma: 0.117    huber_delta: 0.118    log_every: 10019    eval_every: 25020    seq_len: 51221  lambda_grid:22  - 0.023  - 0.0124  - 0.0325  - 0.126  - 0.327  kappa_grid:28  - 0.029  - 0.530  sweep_batches: 631  max_budget: 832  selection_rule: max_loss_reduction_per_ops33  model_overrides: {}34model:35  name: prisme_103m36  vocab_size: 3276837  d_model: 76838  n_layers: 1239  n_heads: 1240  n_kv_heads: 341  ffn_width: 204842  max_seq_len: 409643  rope_theta: 10000.044  norm_eps: 1.0e-0545  tie_embeddings: true46  init_std: 0.0247  use_workspace: true48  w: 25649  ffn_w: 76850  n_slots: 1651  n_editor_blocks: 252  n_editor_heads: 453  r_max: 854  eta_max: 0.2555  clip_c: 1.056  clip_eps: 1.0e-0657  readout_zero_init: false58  readout_init_std: 0.259  readout_scale: 8.85156631469726660  controller_hidden: 061  ctrl_input: h62  ctrl_use_entropy: true63  lambda_cost: 0.064  kappa_risk: 0.065  dense_writes: false66
Files · main
Menu

Theme