{
  "run": "baseline_en",
  "step": 5000,
  "model_config": {
    "vocab_size": 256,
    "d_enc": 256,
    "d_main": 512,
    "n_enc": 2,
    "n_dec": 2,
    "n_main": 6,
    "n_heads_enc": 4,
    "n_heads_main": 8,
    "seq_len": 1024,
    "dropout": 0.0,
    "target_ratio": 6.0,
    "max_chunk_mult": 3.0,
    "dechunk": "ema",
    "rope_base": 10000.0
  },
  "tied_weights": {
    "head.weight": "emb.weight"
  },
  "train_config": {
    "condition": "A",
    "seed": 0,
    "out_dir": null,
    "run_name": "baseline_en",
    "corpus": "standin",
    "langs": [
      "en"
    ],
    "seq_len": 1024,
    "batch_size": 32,
    "grad_accum": 1,
    "steps": 5000,
    "lr": 0.0003,
    "min_lr_frac": 0.1,
    "warmup": 200,
    "weight_decay": 0.1,
    "grad_clip": 1.0,
    "betas": [
      0.9,
      0.95
    ],
    "router_lr_mult": 1.0,
    "d_enc": 256,
    "d_main": 512,
    "n_enc": 2,
    "n_dec": 2,
    "n_main": 6,
    "n_heads_enc": 4,
    "n_heads_main": 8,
    "dechunk": "ema",
    "target_ratio": 6.0,
    "alpha": 0.03,
    "alpha_parity_cap": 0.06,
    "global_anchor": 0.0,
    "premium_clip": 4.0,
    "premium_kind": "per_char",
    "tau1": 0.4,
    "tau2": 0.7,
    "freeze_scope": "router",
    "ckpt_every": 500,
    "eval_every": 250,
    "log_every": 50,
    "probe_sents": 96,
    "eval_batches": 8,
    "keep_ckpt": false,
    "amp": "bf16",
    "compile": false,
    "cpc_decay": 0.9,
    "device": "cuda"
  },
  "probe_agg": {
    "gini_cpc": 0.0,
    "gini_cps": 0.0,
    "bpic": 5.065967016491754,
    "worst_lang": "en",
    "cpc_parity_vs_en": {
      "en": 1.0
    }
  },
  "source": "AdaptiveChunking/hnet-chunking-pilots/baseline_en/model.safetensors"
}