Hugging Face's logo

AbstractPhil
/
geolip-bytelex

tokenization
byte-level
distillation
lexicon
Model card Files
xet
 Community
geolip-bytelex
File size: 2,335 Bytes
ba6a2b4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
 "instrument": "btx-e003 R0b I-5 codebooks (model-free recipe)",
 "when": "2026-09-06 00:47",
 "corpora": {
  "d5": {
   "n_states": 27,
   "dims_used": 27,
   "class_counts": {
    "word": 18,
    "Name": 9,
    "digit": 0
   },
   "count_1_dropped": 0,
   "top_states": [
    [
     "is",
     264000
    ],
    [
     "If",
     120000
    ],
    [
     "So",
     120000
    ],
    [
     "are",
     120000
    ],
    [
     "someone",
     120000
    ],
    [
     "then",
     120000
    ],
    [
     "they",
     120000
    ],
    [
     "tovi",
     32270
    ],
    [
     "marn",
     32175
    ],
    [
     "trell",
     32173
    ],
    [
     "dorn",
     32077
    ],
    [
     "vint",
     32048
    ]
   ],
   "singular_share_top4": 0.5727,
   "erank_feats": 11.9,
   "rsa_E0_vs_raw_feature_cosines": 0.0653,
   "rsa_convention_note": "the record gauge RSA(E_C, byte features) compares a TRAINED codebook against E0 (its init); this row is E0 against the raw centered feature cosines (low by construction: the SVD init equalizes directions)",
   "rsa_E0_vs_random": -0.015,
   "b_prior_range": [
    -4.188,
    -1.776
   ]
  },
  "sub3": {
   "n_states": 1009,
   "dims_used": 256,
   "class_counts": {
    "word": 6,
    "Name": 4,
    "digit": 999
   },
   "count_1_dropped": 0,
   "top_states": [
    [
     "is",
     40100
    ],
    [
     "Compute",
     20050
    ],
    [
     "It",
     20050
    ],
    [
     "Reply",
     20050
    ],
    [
     "What",
     20050
    ],
    [
     "minus",
     20050
    ],
    [
     "number",
     20050
    ],
    [
     "only",
     20050
    ],
    [
     "the",
     20050
    ],
    [
     "with",
     20050
    ],
    [
     "122",
     220
    ],
    [
     "135",
     206
    ]
   ],
   "singular_share_top4": 0.3015,
   "erank_feats": 50.5,
   "rsa_E0_vs_raw_feature_cosines": 0.096,
   "rsa_convention_note": "the record gauge RSA(E_C, byte features) compares a TRAINED codebook against E0 (its init); this row is E0 against the raw centered feature cosines (low by construction: the SVD init equalizes directions)",
   "rsa_E0_vs_random": -0.0009,
   "b_prior_range": [
    -9.156,
    -2.14
   ]
  }
 }
}