Hugging Face's logo

AdaptiveChunking
/
parity_B_ste_s0

PyTorch
Safetensors
dynamic-chunking
tokenization
hnet
byte-level
beetle-hnet
Model card Files
xet
 Community
parity_B_ste_s0
File size: 3,402 Bytes
37807f1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
{
  "arch": "hnet",
  "step": 6000,
  "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": "ste",
    "router_noise": 0.0,
    "rope_base": 10000.0
  },
  "ali_config": null,
  "tied_weights": {
    "head.weight": "emb.weight"
  },
  "config": {
    "model": {
      "architecture": "hnet",
      "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": "ste",
      "router_noise": 0.0,
      "rope_base": 10000.0
    },
    "memory": {
      "ali": {
        "budget_B": 65536,
        "staging_capacity": 262144,
        "policy": "utility",
        "candidate_types": [
          "surface",
          "split",
          "merge",
          "skel"
        ],
        "write_rule": "mean",
        "eta_min": 0.0,
        "memory_scope": "persistent",
        "consolidate_every": 50,
        "max_span": 64,
        "gate": "vector",
        "retrieval": true,
        "read_dropout": 0.1,
        "gate_hidden": 0,
        "type_dim": 16,
        "gate_bias_init": -2.0,
        "probe_seqs": 2,
        "probe_reps": 8,
        "horizon": 32,
        "predictor": "mlp",
        "lambda_util": 0.1,
        "enabled": true,
        "inventory_kwargs": {},
        "probe_mix": null,
        "util_batch_slots": 1024
      },
      "pkm": {
        "n_keys": 256,
        "topk": 8,
        "heads": 1,
        "d_key": null,
        "value_lr_mult": 1.0
      }
    },
    "objective": {
      "name": "parity",
      "schedule": "B",
      "alpha": 0.03,
      "alpha_parity_cap": 0.06,
      "global_anchor": 0.0,
      "premium_clip": 4.0,
      "premium_kind": "flores_per_sent",
      "tau1": 0.4,
      "tau2": 0.7,
      "freeze_scope": "router",
      "cpc_decay": 0.9,
      "genreg_target_entropy": null,
      "genreg_eta": 0.05
    },
    "data": {
      "corpus": "corpus",
      "langs": null,
      "mixture": null,
      "doc_filters": null,
      "byte_classes": "scratch/data/byte_classes.json"
    },
    "training": {
      "seed": 0,
      "max_steps": 6000,
      "batch_size": 32,
      "grad_accum": 1,
      "lr": 0.0003,
      "min_lr_frac": 0.1,
      "warmup": 300,
      "weight_decay": 0.1,
      "grad_clip": 1.0,
      "betas": [
        0.9,
        0.95
      ],
      "router_lr_mult": 1.0,
      "log_every": 50,
      "amp": "bf16",
      "compile": false,
      "device": "cuda"
    },
    "checkpointing": {
      "run_name": "parity_B_ste_s0",
      "runs_dir": "runs",
      "ckpt_every": 200,
      "keep": false,
      "milestones": [
        1000,
        3000
      ],
      "resume": "auto"
    },
    "evaluation": {
      "eval_every": 200,
      "eval_batches": 8,
      "probe_sents": 128,
      "probe_bpb_batches": 4,
      "log_spaced_probe": true,
      "dense_early_every": 0,
      "dense_early_until": 0,
      "dump_reps": true,
      "dump_reps_thresholds": [
        0.3,
        0.4,
        0.5,
        0.6,
        0.7
      ]
    }
  },
  "format": "beetle-hnet step branch v1 (src/checkpointing/hub.py)"
}