File size: 10,830 Bytes
3a14f46
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
{
  "bundle_format": "foldmatch-esm-codes-bundle/1",
  "system": "esm-codes-v1",
  "calibration_version": "foldmatch-evalue-M8_s0-D_fmr-pnlU50s1000-M3t999-v1",
  "hf_repo": "rcsb/foldmatch-esm-codes",
  "files": {
    "esm_codes_v1_encoders.pt": {
      "sha256": "1696ab91089ece8bce787d38b9d4d68bbc8ca4abff06f836f6e9363a23a1c02b",
      "bytes": 23324662,
      "format": "foldmatch-esm-codes-encoders/1",
      "load": "torch.load(path, map_location='cpu', weights_only=True)"
    },
    "esm_codes_v1_scoring.npz": {
      "sha256": "a5bebd626ec874626fb9372a63d10c32a001ed09c358717448bc9e1fd0ddedd6",
      "bytes": 2983002,
      "format": "foldmatch-esm-codes-scoring/1",
      "load": "numpy.load(path, allow_pickle=False)"
    }
  },
  "content_digests": {
    "method": "foldmatch-tensor-digest/1",
    "method_detail": "sha256 over the ASCII tag 'foldmatch-tensor-digest/1' and a newline, then for each array name in sorted order: an 8-byte little-endian length and the compact sorted-key JSON {dtype (numpy dtype str, little-endian), name, shape}, an 8-byte little-endian length and the raw C-order little-endian bytes",
    "transform_sha256": "e8989b1cefc839784555ff973dabb9a2f7ef4bdf81ac78ad47767714227c3336",
    "query_encoder_sha256": "d8bed195068be23a60586632fa471a08f3a767fe8cf34b06aa6e733d90cd3bf8",
    "target_encoder_sha256": "58da263e005a79fd1689b0f5605adbb85b51532bca99e4ef7330916a50df6686",
    "transform_arrays": [
      "M",
      "dims",
      "offset"
    ],
    "encoder_arrays": [
      "fc1.bias",
      "fc1.weight",
      "fc2.bias",
      "fc2.weight",
      "lin.bias",
      "lin.weight"
    ],
    "panel_arrays": [
      "codes (panel_codes)",
      "offsets (panel_offsets)"
    ],
    "table_arrays": [
      "bin_centres",
      "lam",
      "lnK"
    ],
    "panel_sha256": "7aea2b00a24ee41c5a9f9adb91a5723ee02dff12dca23998767ab87da4b9dcf7",
    "table_sha256": "67bc270c1c3281822a9bed58bc645086327a9bd315e38f53f6439a80b1469ae9"
  },
  "esm3": {
    "model": "esm3_sm_open_v1",
    "repo": "rcsb/rcsb-esm",
    "revision": "636acb420daf674df37ab0220f56061a5de2f169",
    "weights_file": "data/weights/esm3_sm_open_v1.pth",
    "weights_sha256": "5ead5a135c658068db6a4f1b933e72d6110992c4668822e1c0e2dcc53e38acd9",
    "upstream_repo": "EvolutionaryScale/esm3-sm-open-v1",
    "upstream_revision": "47f0545b2b6daf26a93439a3cd610f4f7f3d5478",
    "embedding_tap": "per_residue_embedding",
    "embedding_tap_detail": "ESM3.forward_and_sample(..., SamplingConfig(return_per_residue_embeddings=True)).per_residue_embedding: the last transformer block's output before the final LayerNorm; rows 1..L (BOS and EOS removed), [L, 1536]; sequence-only input",
    "input_normalisation": "ascii_upper_nows_v1",
    "input_normalisation_detail": "ASCII whitespace removed, ASCII letters upper-cased, non-ASCII refused (S15)"
  },
  "encoding": {
    "architecture": {
      "name": "esm-codes-mlp-v1",
      "d_embed": 1536,
      "dims_dropped": [
        1361,
        1489
      ],
      "d_in": 1534,
      "hidden": 1024,
      "d_out": 64,
      "activation": "gelu_exact",
      "skip": "linear",
      "bias": true,
      "dropout": 0.0,
      "encoders": [
        "f_q",
        "f_t"
      ]
    },
    "transform": "x16 = fp16(x); xs = fp32(x16[:, dims]) - offset; xw = fp32(fp16(xs @ M))",
    "input_cast": "float16",
    "transform_output_cast": "float16",
    "encoder": {
      "formula": "f(x) = lin(x) + fc2(GELU_exact(fc1(x)))",
      "dtype": "float32",
      "tf32": false,
      "autocast": false,
      "dropout": 0.0
    },
    "code_bits": 64,
    "code_bytes": 8,
    "bitorder": "big",
    "sign_rule": "bit_k = (h_k >= 0)",
    "query_dtype": "float16"
  },
  "scoring": {
    "dp": "fmsw/1",
    "offset": 1.0,
    "gap_open": 5.0,
    "gap_extend": 0.5,
    "gap_cost": "a gap of length L costs gap_open + (L - 1) * gap_extend",
    "cell_score": "z_ij - offset",
    "score_form": "fwd_minus_rev_view",
    "score_form_detail": "S = fp64(S_f) - fp64(S_r); S_r = Smith-Waterman on the row-reversed view of the same similarity, z recomputed in its own row order, score only",
    "score_diff": "float64",
    "normalisation": {
      "name": "rowcol_z",
      "std": "population",
      "eps": 1e-08,
      "degenerate_std_lt": 1e-06,
      "degenerate_rule": "a row (column) with std < 1e-6 contributes exactly 0",
      "formula": "z_ij = 0.5 ((s_ij - mu_i) / (sigma_i + 1e-8) + (s_ij - nu_j) / (tau_j + 1e-8))"
    },
    "query_prep": "fp64_seq_k_asc",
    "query_prep_detail": "qp_i = fp32(fp64(q_i) / (8 sqrt(sum_k fp64(q_ik)^2))), k ascending; |q_i| < 1e-12 -> 0; NaN / Inf is an error",
    "query_prep_eps": 1e-12,
    "similarity": "pm1_fp32_blas_gemm_per_pair",
    "determinism": "tolerance_s16_v1"
  },
  "statistics": {
    "table_key": "D_fmr/pnl_uniref50/M3t999",
    "bin_centres": [
      32.58437573116228,
      50.34161310783617,
      80.8700900525453,
      149.85910754690784,
      276.49344608131264,
      594.9175724688861
    ],
    "lam": [
      0.18775162858061797,
      0.1227859873586807,
      0.10488523441851651,
      0.08222582327223689,
      0.04784353519669357,
      0.041619283794126286
    ],
    "lnK": [
      -13.42382627313042,
      -13.969792295008805,
      -13.73906582316063,
      -13.454607332784638,
      -14.484498724905178,
      -15.878915750325046
    ],
    "interpolation": "lam(m), lnK(m) linear in ln m between the six bin centres (numpy.interp(ln m, ln centres, table))",
    "low_m_rule": "clamp",
    "high_m_rule": "clamp",
    "m": "query length (residues)",
    "N": "database residues (manifest total_residues)",
    "S_prime": "S - Q90(query)",
    "lnE": "lnK(m) + ln m + ln N - lam(m) * S'  (float64)",
    "bits": "(lam(m) * S' - lnK(m)) / ln 2, rendered floor(bits + 0.5)",
    "fit": "real-null censored Gumbel ML (rlib.fit_m3, ship.py procedure), q .999, bins 0-40/40-60/60-100/100-200/200-400/400+, bin centre = geometric-mean query length; 1,465 benchmark queries x all 50,000 targets (real null = annotated targets with no shared SCOP fold, evaluator ignore rules)",
    "fit_db_residues": 8557236.0
  },
  "panel": {
    "size": 1000,
    "residues": 355626,
    "quantile": 0.9,
    "quantile_method": "linear",
    "q90": "numpy.quantile(float64 S(query, members), 0.9), linear interpolation",
    "source": "UniRef50 release 2026_03 cluster representatives; AMENDED stratified draw 0 of size 1000 (panel/amend/AMENDMENT.txt md5 589708cc06a710d8007c7161190e175c, draws.tsv md5 3e726d11361cae298582b3680c97f0de); S* from select2/results/decision.json (A1 rule)",
    "codes_bytes_md5": "885dfc88590e0b270addcd4b61c3a7a7",
    "idlist_md5": {
      "rank_order": "d416d3b74222a7ffb65dae5cc7d4eee6",
      "sorted": "539809b6758074294977b38b15486008",
      "ids_file": "68463819ee52018dccb6b4fab3d5467a"
    },
    "panel_sha256": "7aea2b00a24ee41c5a9f9adb91a5723ee02dff12dca23998767ab87da4b9dcf7"
  },
  "calibrated_domain": {
    "query_min_length": 21,
    "query_max_length": 966,
    "target_max_length": 3000,
    "max_nonstandard_fraction": 0.01,
    "target_min_distinct_codes": 2,
    "nonstandard": "any character outside the 20 standard amino acids ACDEFGHIKLMNPQRSTVWY",
    "rule": "ecal = 1 only when every condition holds; otherwise ecal = 0, the alignment is still reported and a once-per-run warning gives the counts"
  },
  "degenerate_query_rule": "no_hits",
  "degenerate_query_detail": "a query with n = 1 or whose q fp16 rows are all bitwise identical reports no hits (S13b), whatever max_evalue",
  "licence": {
    "spdx": "MIT",
    "file": "LICENSE",
    "copyright": "Copyright (c) 2026 - now, RCSB PDB and contributors",
    "scope": "MIT: the encoders, the zcw50_clean transform, the statistics table, the panel codes and the documentation. Not covered by the MIT licence: the panel member identifiers (panel_ids), which are UniProt data under CC BY 4.0 (see attribution.uniref50).",
    "exceptions": {
      "panel_ids": {
        "spdx": "CC-BY-4.0",
        "source": "UniProt Consortium, UniRef50 release 2026_03",
        "attribution": "attribution.uniref50"
      }
    },
    "attribution": {
      "uniref50": "Panel member identifiers are UniRef50 cluster identifiers (UniProt release 2026_03) from the UniProt Consortium, https://www.uniprot.org, distributed under the Creative Commons Attribution 4.0 International licence (CC BY 4.0, https://creativecommons.org/licenses/by/4.0/). The panel codes were computed from ESM3 embeddings of those representative sequences; the sequences themselves are not included.",
      "esm3": "The encoders take per-residue embeddings of ESM3 esm3_sm_open_v1 (EvolutionaryScale) as input; no ESM3 weights are included. FoldMatch loads the weights from rcsb/rcsb-esm@636acb420daf674df37ab0220f56061a5de2f169 (MIT; weights sha256 5ead5a135c658068db6a4f1b933e72d6110992c4668822e1c0e2dcc53e38acd9), the same blob as EvolutionaryScale/esm3-sm-open-v1@47f0545b2b6daf26a93439a3cd610f4f7f3d5478, whose model card points to the MIT licence of the evolutionaryscale/esm repository. Older releases of the same weights (e.g. the gated EvolutionaryScale/esm3-sm-open-v1 card at 2feda652 and the LICENSE.md shipped in the esm 3.2.3 wheel) carry the EvolutionaryScale Cambrian Non-Commercial License; each copy of the weights is governed by the licence it was obtained under."
    }
  },
  "provenance": {
    "research_model": "L2m_B8_s0 (sw_mlp), checkpoint epoch 7",
    "inputs": {
      "encoders": {
        "file": "L2m_B8_s0.pt",
        "md5": "85fdd985087d6813e9994274d4fca6ab",
        "bytes": 13891299
      },
      "encoders_history": {
        "file": "L2m_B8_s0.json",
        "md5": "19e58f8956cf24404def09aac56e00b5",
        "bytes": 50732
      },
      "sw_common": {
        "file": "sw_common.py",
        "md5": "b8dab0a03907c9b7ed67663d96c55561",
        "bytes": 12622
      },
      "zcw50_clean": {
        "file": "zcw50_clean.npz",
        "md5": "4c7692ed66e599aca906b41ec7c13f04",
        "bytes": 37689630
      },
      "ship_final": {
        "file": "ship_final.json",
        "md5": "c51735f962c04bd8c441b9625c4cf30a",
        "bytes": 36685
      },
      "panel_codes": {
        "file": "shipped_panel_codes.npz",
        "md5": "15cc1c7b2d1dd488a5f9b742365ccb13",
        "bytes": 8020960
      },
      "panel_ids": {
        "file": "shipped_panel.ids",
        "md5": "68463819ee52018dccb6b4fab3d5467a",
        "bytes": 18536
      }
    },
    "generator": {
      "script": "rcsb-embedding-search src/scripts/save_esm_codes_for_production.py",
      "sha256": "722eece5fdffb2c8ba2d0bd36ef01a4ac6d8ca3762301709189f52e43482920c"
    },
    "environment": {
      "python": "3.12.12",
      "torch": "2.9.0",
      "numpy": "2.4.4",
      "platform": "Darwin-arm64"
    }
  }
}