fengzhimu kierandidi commited on
Commit
a40d48a
·
0 Parent(s):

Duplicate from nvidia/RFD4-Proteina

Browse files

Co-authored-by: Kieran Didi <kierandidi@users.noreply.huggingface.co>

This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +219 -0
  2. LICENSE +28 -0
  3. README.md +309 -0
  4. adapter/adapter_config.json +46 -0
  5. adapter/adapter_model.safetensors +3 -0
  6. ae_checkpoints/rfd4_proteina_ae-ema.ckpt +3 -0
  7. ae_checkpoints/rfd4_proteina_ae-ema.ckpt.sha256 +1 -0
  8. alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sif +3 -0
  9. alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sif.sha256 +1 -0
  10. alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sqsh +3 -0
  11. alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sqsh.sha256 +1 -0
  12. alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sif +3 -0
  13. alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sif.sha256 +1 -0
  14. alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sqsh +3 -0
  15. alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sqsh.sha256 +1 -0
  16. assets/design-gallery.png +3 -0
  17. assets/examples/antibody_design.png +3 -0
  18. assets/examples/enzyme_design.png +3 -0
  19. assets/examples/protein_binder.png +3 -0
  20. assets/wordmark-dark.webp +0 -0
  21. assets/wordmark.webp +0 -0
  22. base/config.json +263 -0
  23. base/model.safetensors +3 -0
  24. benchmarks/METHODS.md +56 -0
  25. benchmarks/provenance.json +13 -0
  26. benchmarks/source_data.csv +0 -0
  27. benchmarks/summary.csv +53 -0
  28. config.json +263 -0
  29. examples/antibody_design/design.cif +0 -0
  30. examples/antibody_design/input.cif +0 -0
  31. examples/enzyme_design/design.cif +0 -0
  32. examples/enzyme_design/input.cif +177 -0
  33. examples/monomer.cif +78 -0
  34. examples/protein_binder/design.cif +0 -0
  35. examples/protein_binder/input.cif +0 -0
  36. flow_checkpoints/rfd4_proteina_flow-ema.ckpt +3 -0
  37. flow_checkpoints/rfd4_proteina_flow-ema.ckpt.sha256 +1 -0
  38. flow_checkpoints/rfd4_proteina_flow_pretrain-ema.ckpt +3 -0
  39. flow_checkpoints/rfd4_proteina_flow_pretrain-ema.ckpt.sha256 +1 -0
  40. inference.yaml +79 -0
  41. model.safetensors +3 -0
  42. openfold3/runtime/of3_af3_refolding_env-cuda12.docker.tar +3 -0
  43. openfold3/runtime/of3_af3_refolding_env-cuda12.docker.tar.sha256 +3 -0
  44. openfold3/runtime/of3_af3_refolding_env-cuda12.sif +3 -0
  45. openfold3/runtime/of3_af3_refolding_env-cuda12.sif.sha256 +1 -0
  46. openfold3/runtime/of3_af3_refolding_env-cuda12.sqsh +3 -0
  47. openfold3/runtime/of3_af3_refolding_env-cuda12.sqsh.sha256 +1 -0
  48. openfold3/runtime/of3_af3_refolding_env-cuda13.docker.tar +3 -0
  49. openfold3/runtime/of3_af3_refolding_env-cuda13.docker.tar.sha256 +3 -0
  50. openfold3/runtime/of3_af3_refolding_env-cuda13.sif +3 -0
.gitattributes ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ af3/apptainer_images/af3-open-36fad16.sif filter=lfs diff=lfs merge=lfs -text
37
+ aoti/h200/b1-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
38
+ aoti/h200/b2-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
39
+ aoti/h200/b4-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
40
+ aoti/h200/b8-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
41
+ aoti/h200/b1-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
42
+ aoti/h200/b1-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
43
+ aoti/h200/b1-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
44
+ aoti/h200/b1-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
45
+ aoti/blackwell/b1-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
46
+ aoti/blackwell/b2-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
47
+ aoti/blackwell/b4-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
48
+ aoti/blackwell/b8-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
49
+ aoti/blackwell/b1-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
50
+ aoti/blackwell/b1-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
51
+ aoti/pasco-blackwell/b1-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
52
+ aoti/pasco-blackwell/b1-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
53
+ aoti/pasco-blackwell/b1-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
54
+ aoti/pasco-blackwell/b1-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
55
+ aoti/pasco-blackwell/b1-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
56
+ aoti/pasco-blackwell/b1-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
57
+ aoti/pasco-blackwell/b1-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
58
+ aoti/pasco-blackwell/b1-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
59
+ aoti/pasco-blackwell/b1-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
60
+ aoti/pasco-blackwell/b1-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
61
+ aoti/pasco-blackwell/b1-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
62
+ aoti/pasco-blackwell/b1-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
63
+ aoti/pasco-blackwell/b1-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
64
+ aoti/pasco-blackwell/b1-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
65
+ aoti/pasco-blackwell/b1-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
66
+ aoti/pasco-blackwell/b2-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
67
+ aoti/pasco-blackwell/b4-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
68
+ aoti/pasco-blackwell/b8-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
69
+ aoti/pasco-blackwell/b2-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
70
+ aoti/pasco-blackwell/b4-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
71
+ aoti/pasco-blackwell/b8-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
72
+ aoti/pasco-blackwell/b2-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
73
+ aoti/pasco-blackwell/b4-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
74
+ aoti/pasco-blackwell/b8-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
75
+ aoti/pasco-blackwell/b2-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
76
+ aoti/pasco-blackwell/b4-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
77
+ aoti/pasco-blackwell/b8-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
78
+ aoti/pasco-blackwell/b2-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
79
+ aoti/pasco-blackwell/b4-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
80
+ aoti/pasco-blackwell/b8-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
81
+ openfold3/runtime/openfold3-36fad16.sif filter=lfs diff=lfs merge=lfs -text
82
+ assets/examples/protein_binder.png filter=lfs diff=lfs merge=lfs -text
83
+ assets/examples/enzyme_design.png filter=lfs diff=lfs merge=lfs -text
84
+ assets/examples/antibody_design.png filter=lfs diff=lfs merge=lfs -text
85
+ aoti/l40s/b1-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
86
+ aoti/l40s/b1-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
87
+ aoti/l40s/b1-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
88
+ aoti/l40s/b1-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
89
+ aoti/l40s/b1-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
90
+ aoti/l40s/b1-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
91
+ aoti/l40s/b1-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
92
+ aoti/l40s/b1-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
93
+ aoti/a100/b1-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
94
+ aoti/a100/b1-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
95
+ aoti/a100/b1-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
96
+ aoti/a100/b1-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
97
+ aoti/a100/b1-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
98
+ aoti/a100/b1-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
99
+ aoti/a100/b1-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
100
+ aoti/a100/b1-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
101
+ aoti/l40s/b1-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
102
+ aoti/l40s/b1-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
103
+ aoti/l40s/b1-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
104
+ aoti/l40s/b1-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
105
+ aoti/l40s/b1-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
106
+ aoti/l40s/b1-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
107
+ aoti/l40s/b1-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
108
+ aoti/l40s/b2-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
109
+ aoti/l40s/b2-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
110
+ aoti/l40s/b2-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
111
+ aoti/l40s/b2-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
112
+ aoti/l40s/b2-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
113
+ aoti/l40s/b2-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
114
+ aoti/l40s/b2-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
115
+ aoti/l40s/b2-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
116
+ aoti/l40s/b2-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
117
+ aoti/l40s/b2-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
118
+ aoti/l40s/b2-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
119
+ aoti/l40s/b2-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
120
+ aoti/l40s/b2-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
121
+ aoti/l40s/b2-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
122
+ aoti/l40s/b2-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
123
+ aoti/l40s/b4-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
124
+ aoti/l40s/b4-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
125
+ aoti/l40s/b4-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
126
+ aoti/l40s/b4-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
127
+ aoti/l40s/b4-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
128
+ aoti/l40s/b4-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
129
+ aoti/l40s/b4-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
130
+ aoti/l40s/b4-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
131
+ aoti/l40s/b4-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
132
+ aoti/l40s/b4-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
133
+ aoti/l40s/b4-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
134
+ aoti/l40s/b4-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
135
+ aoti/l40s/b4-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
136
+ aoti/l40s/b4-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
137
+ aoti/l40s/b4-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
138
+ aoti/l40s/b8-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
139
+ aoti/l40s/b8-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
140
+ aoti/l40s/b8-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
141
+ aoti/l40s/b8-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
142
+ aoti/l40s/b8-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
143
+ aoti/l40s/b8-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
144
+ aoti/l40s/b8-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
145
+ aoti/l40s/b8-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
146
+ aoti/l40s/b8-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
147
+ aoti/l40s/b8-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
148
+ aoti/l40s/b8-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
149
+ aoti/l40s/b8-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
150
+ aoti/l40s/b8-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
151
+ aoti/a100/b1-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
152
+ aoti/a100/b1-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
153
+ aoti/a100/b1-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
154
+ aoti/a100/b1-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
155
+ aoti/a100/b1-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
156
+ aoti/a100/b1-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
157
+ aoti/a100/b1-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
158
+ aoti/a100/b2-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
159
+ aoti/a100/b2-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
160
+ aoti/a100/b2-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
161
+ aoti/a100/b2-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
162
+ aoti/a100/b2-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
163
+ aoti/a100/b2-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
164
+ aoti/a100/b2-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
165
+ aoti/a100/b2-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
166
+ aoti/a100/b2-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
167
+ aoti/a100/b2-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
168
+ aoti/a100/b2-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
169
+ aoti/a100/b2-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
170
+ aoti/a100/b2-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
171
+ aoti/a100/b2-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
172
+ aoti/a100/b2-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
173
+ aoti/a100/b4-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
174
+ aoti/a100/b4-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
175
+ aoti/a100/b4-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
176
+ aoti/a100/b4-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
177
+ aoti/a100/b4-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
178
+ aoti/a100/b4-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
179
+ aoti/a100/b4-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
180
+ aoti/a100/b4-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
181
+ aoti/a100/b4-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
182
+ aoti/a100/b4-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
183
+ aoti/a100/b4-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
184
+ aoti/a100/b4-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
185
+ aoti/a100/b4-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
186
+ aoti/a100/b4-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
187
+ aoti/a100/b4-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
188
+ aoti/a100/b8-n1024/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
189
+ aoti/a100/b8-n128/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
190
+ aoti/a100/b8-n160/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
191
+ aoti/a100/b8-n192/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
192
+ aoti/a100/b8-n224/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
193
+ aoti/a100/b8-n256/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
194
+ aoti/a100/b8-n288/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
195
+ aoti/a100/b8-n320/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
196
+ aoti/a100/b8-n352/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
197
+ aoti/a100/b8-n384/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
198
+ aoti/a100/b8-n448/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
199
+ aoti/a100/b8-n512/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
200
+ aoti/a100/b8-n640/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
201
+ aoti/a100/b8-n768/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
202
+ aoti/a100/b8-n896/root/package.pt2 filter=lfs diff=lfs merge=lfs -text
203
+ assets/design-gallery.png filter=lfs diff=lfs merge=lfs -text
204
+ openfold3/runtime/of3_af3_refolding_env.sif filter=lfs diff=lfs merge=lfs -text
205
+ alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sif filter=lfs diff=lfs merge=lfs -text
206
+ alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sqsh filter=lfs diff=lfs merge=lfs -text
207
+ openfold3/runtime/of3_af3_refolding_env.sqsh filter=lfs diff=lfs merge=lfs -text
208
+ openfold3/runtime/of3_af3_refolding_env-cuda12.sif filter=lfs diff=lfs merge=lfs -text
209
+ openfold3/runtime/of3_af3_refolding_env-cuda12.sqsh filter=lfs diff=lfs merge=lfs -text
210
+ alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sif filter=lfs diff=lfs merge=lfs -text
211
+ alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sqsh filter=lfs diff=lfs merge=lfs -text
212
+ openfold3/runtime/of3_af3_refolding_env-cuda13.sif filter=lfs diff=lfs merge=lfs -text
213
+ openfold3/runtime/of3_af3_refolding_env-cuda13.sqsh filter=lfs diff=lfs merge=lfs -text
214
+ runtime/rfd4-proteina-cpu-amd64.sif filter=lfs diff=lfs merge=lfs -text
215
+ runtime/rfd4-proteina-cpu-amd64.sqsh filter=lfs diff=lfs merge=lfs -text
216
+ runtime/rfd4-proteina-cu12-amd64.sif filter=lfs diff=lfs merge=lfs -text
217
+ runtime/rfd4-proteina-cu12-amd64.sqsh filter=lfs diff=lfs merge=lfs -text
218
+ runtime/rfd4-proteina-cu13-amd64.sif filter=lfs diff=lfs merge=lfs -text
219
+ runtime/rfd4-proteina-cu13-amd64.sqsh filter=lfs diff=lfs merge=lfs -text
LICENSE ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ BSD 3-Clause License
2
+
3
+ Copyright (c) 2026, Institute for Protein Design, University of Washington
4
+
5
+ Redistribution and use in source and binary forms, with or without
6
+ modification, are permitted provided that the following conditions are met:
7
+
8
+ * Redistributions of source code must retain the above copyright notice, this
9
+ list of conditions and the following disclaimer.
10
+
11
+ * Redistributions in binary form must reproduce the above copyright notice,
12
+ this list of conditions and the following disclaimer in the documentation
13
+ and/or other materials provided with the distribution.
14
+
15
+ * Neither the name of the copyright holder nor the names of its
16
+ contributors may be used to endorse or promote products derived from
17
+ this software without specific prior written permission.
18
+
19
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20
+ AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21
+ IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
23
+ FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24
+ DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25
+ SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26
+ CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
27
+ OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28
+ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
README.md ADDED
@@ -0,0 +1,309 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: bsd-3-clause
3
+ library_name: rfd4-proteina
4
+ language:
5
+ - en
6
+ datasets:
7
+ - nvidia/RFD4-Proteina
8
+ tags:
9
+ - biology
10
+ - protein-design
11
+ - biomolecular-design
12
+ - structure-generation
13
+ - flow-matching
14
+ - annotated-cif
15
+ ---
16
+ <p align="center"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/wordmark-dark.webp"><img src="assets/wordmark.webp" width="520" alt="RFD4-Proteína"></picture></p>
17
+ <p align="center"><strong>Sequence and structure, designed together.</strong><br>Baker Lab × NVIDIA GenAIR</p>
18
+ <p align="center"><a href="https://rfd4-proteina.org/">Project page</a> · <a href="https://github.com/RosettaCommons/RFD4-Proteina">Code</a> · <a href="https://rfd4-proteina.org/latest/">Documentation</a> · <a href="https://huggingface.co/datasets/nvidia/RFD4-Proteina">Training data</a> · <a href="https://huggingface.co/spaces/nvidia/rfd4-proteina-zerogpu">Studio (ZeroGPU)</a> · <a href="https://huggingface.co/spaces/nvidia/rfd4-proteina-space">Studio (dedicated GPU)</a></p>
19
+
20
+ # RFD4-Proteína
21
+
22
+ RFD4-Proteína is a flow-matching model for biomolecular design. It generates protein sequences and all-atom structures together, conditioned on the molecular context you want to preserve: a binding target, an enzyme motif, a ligand, or selected structural features.
23
+
24
+ <img src="assets/design-gallery.png" width="1050" alt="Examples from the project: nanobody complex, protein–DNA binder, RNA pseudoknot, ligand pocket, glycan complex, and protein assembly.">
25
+
26
+ ## Where to find each entry point
27
+
28
+ This repository is the **canonical native bundle**. `config.json` and `model.safetensors` at
29
+ the root are the *native* weights, loaded by `Proteina.from_pretrained`.
30
+
31
+ | Call | Weights | Location | Status |
32
+ |---|---|---|---|
33
+ | `Proteina.from_pretrained("nvidia/RFD4-Proteina")` | Native | root of this repo | **Supported** |
34
+ | `RFD4ProteinaPipeline.from_pretrained("nvidia/RFD4-Proteina-Diffusers")` | Diffusers | [its own repo](https://huggingface.co/nvidia/RFD4-Proteina-Diffusers) | **Supported**, Diffusers from git until the next release |
35
+
36
+ It is one model throughout — nothing is retrained. The Diffusers weights are the same tensors
37
+ re-prefixed and split across `transformer/` and `vae/`. Diffusers needs `model_index.json` at the
38
+ root of whatever it is pointed at, so it has its own repository. Annotated-CIF preprocessing for
39
+ Diffusers is in [`nvidia/RFD4-Proteina-prep`](https://huggingface.co/nvidia/RFD4-Proteina-prep).
40
+
41
+ > **Diffusers from git until the next release.** `nvidia/RFD4-Proteina-Diffusers` holds the same 2026-10-19
42
+ > release. `RFD4ProteinaPipeline` was merged into Diffusers on 2026-10-20 ([huggingface/diffusers#15088](https://github.com/huggingface/diffusers/pull/15088)) and ships in the next
43
+ > release; until then install Diffusers from GitHub, as in the snippet below.
44
+
45
+ Training data lives in the companion [dataset repository](https://huggingface.co/datasets/nvidia/RFD4-Proteina).
46
+
47
+ ## At a glance
48
+
49
+ | | Details |
50
+ |---|---|
51
+ | **Model** | Joint sequence–structure generation with a matched flow model and autoencoder |
52
+ | **Inputs** | Annotated CIF structures or YAML condition specifications |
53
+ | **Outputs** | Designed molecular structures in CIF format, with generated residue identities and coordinates |
54
+ | **Weights** | EMA weights; 435,622,465 stored parameters across the paired bundle, in FP32 |
55
+ | **Download** | 1.74 GB for `model.safetensors`, plus `config.json`; training data is a separate download |
56
+ | **Runtime** | RFD4 Python package; Linux x86_64, Python 3.12, supported NVIDIA GPU and CUDA 13-compatible driver |
57
+ | **License** | [BSD 3-Clause](LICENSE) for the RFD4 model and weights; optional third-party assets have separate terms |
58
+
59
+ The bundle includes both model components. Use the RFD4 loader below; this is a native RFD4 bundle.
60
+
61
+ ## Checkpoints in this release
62
+
63
+ The release is a **matched pair**: a flow model and the autoencoder it was trained against. They are not interchangeable across releases — the autoencoder defines the latent space the flow model generates in, and mixing a flow model with a different autoencoder produces nonsense rather than an error.
64
+
65
+ | | Current release (2026-10-19) | Previous release |
66
+ |---|---|---|
67
+ | Flow model | `rfd4-proteina-flow-release-2026-10-19`: the pretrained base (step 320000) with a rank-32 LoRA fine-tune merged in | `chk_epoch_00000169_step_000000340000-ema` |
68
+ | Autoencoder | `rfd4-proteina-ae-release-2026-10-19` | `2026_09_04_release_ae_z16_400k-ema` |
69
+ | Latent dimension | 32 | 16 |
70
+ | Residue vocabulary | 147 classes | 33 classes |
71
+ | Stored parameters | 435,622,465 | 434,348,737 |
72
+
73
+ The 2026-10-19 pair supersedes the 2026-10-15 pair (`2026_10_15_hero_s43_c512_80k-ema` and `2026_10_15_unified_ae_s43_400k-ema`), which has the same architecture. The pretrained base alone is in `base/`, and the LoRA adapter that turns it into the release model is in `adapter/` (PEFT format, rank 32 on `to_qkv` and `to_out_node`); the release bundle at the root is exactly the base plus the adapter, merged.
74
+
75
+ The current release's autoencoder is **unified**: one autoencoder covering proteins, nucleic acids and small molecules, with a 147-class residue vocabulary in place of the previous 33-class protein vocabulary, and a 32-dimensional latent in place of 16. This is the change that makes the two releases incompatible with each other's components.
76
+
77
+ | File | Use |
78
+ |---|---|
79
+ | [`config.json`](config.json) + `model.safetensors` | Recommended paired bundle for `Proteina.from_pretrained` |
80
+ | `base/` and `adapter/` | The pretrained base bundle and the PEFT adapter that makes it the release model (`adapter_path=` in the native runtime) |
81
+ | `flow_checkpoints/rfd4_proteina_flow-ema.ckpt` | Native Lightning flow EMA weights |
82
+ | `ae_checkpoints/rfd4_proteina_ae-ema.ckpt` | Matching native Lightning autoencoder EMA weights |
83
+ | [`release-checkpoints.json`](release-checkpoints.json) | Pair identity, stable aliases, source filenames, and SHA-256 checksums |
84
+ | [`provenance.json`](provenance.json) | Source archive and original checkpoint provenance |
85
+
86
+ The published `.ckpt` files carry **EMA weights and hyperparameters only**. The duplicate `SimpleEMA` shadow copy, optimizer, scheduler and loop state are removed, which is why they are 1.74 GB and 1.03 GB rather than 4.21 GB and 3.10 GB. Their weight tensors are bit-identical to the full training checkpoints, whose SHA-256 values are recorded in `release-checkpoints.json`. They initialize weights for inference or fine-tuning; **resuming an interrupted training run needs the full checkpoint**, which is not distributed here.
87
+
88
+ Stable filenames always select the current release; the dated training filename of each is recorded in `release-checkpoints.json`. Pin a Hub revision and verify the SHA-256 sidecars for exact identity. These EMA exports initialize weights; resuming optimizer and scheduler state requires a full training checkpoint.
89
+
90
+ ## Start with one design
91
+
92
+ ### 1. Install
93
+
94
+ Pick one route; the [quickstart](https://rfd4-proteina.org/latest/quickstart.html) covers all three in detail.
95
+
96
+ **GitHub** (recommended: the full workflow with annotated inputs, reward-guided search and refolding metrics):
97
+
98
+ ```bash
99
+ git clone https://github.com/RosettaCommons/RFD4-Proteina.git
100
+ cd RFD4-Proteina
101
+ make setup
102
+ source .venv/bin/activate
103
+ ```
104
+
105
+ **PyPI** (the model inside your own Python environment):
106
+
107
+ ```bash
108
+ python3.12 -m venv .venv && source .venv/bin/activate
109
+ pip install --extra-index-url https://download.pytorch.org/whl/cu132 "rfd4-proteina[metrics-cuda13]"
110
+ ```
111
+
112
+ **Diffusers:** see [Loading the model](#diffusers) below.
113
+
114
+ **Container images** (no build needed): prebuilt RFD4-Proteína runtime images for Linux x86_64 are in [`runtime/`](runtime), as Apptainer (`.sif`) and SquashFS for Enroot/Pyxis (`.sqsh`) files with SHA-256 sidecars. `cu13` (CUDA 13), `cu12` (CUDA 12) and `cpu` (no GPU; imports, CLI and CPU metrics only) variants exist, built from the public [Apptainer recipe](https://github.com/RosettaCommons/RFD4-Proteina/blob/main/apptainer/rfd4-proteina.def). They contain the code and its dependencies, not model weights or the AF2/AF3 oracles; pass `hf_ckpt_path=nvidia/RFD4-Proteina` and a writable `HF_HOME`.
115
+
116
+ ```bash
117
+ hf download nvidia/RFD4-Proteina runtime/rfd4-proteina-cu13-amd64.sif runtime/rfd4-proteina-cu13-amd64.sif.sha256 --local-dir .
118
+ apptainer exec --nv runtime/rfd4-proteina-cu13-amd64.sif rfd4-generate hf_ckpt_path=nvidia/RFD4-Proteina inputs=in.cif out_dir=./runs
119
+ ```
120
+
121
+ Setup caches the paired model bundle; downloads use `HF_HOME` when set. You do not need the training dataset for generation. If access to this repository is restricted, run `hf auth login` first. See [installation](https://rfd4-proteina.org/latest/installation.html) for hardware and build requirements.
122
+
123
+ ### 2. Generate a small protein
124
+
125
+ From the checkout root (with the PyPI install, use the packaged example instead: `inputs=$(python -c "import rfproteina, pathlib; print(pathlib.Path(rfproteina.__file__).parent / 'data/benchmarks/monomer/monomer-short.cif')")`):
126
+
127
+ ```bash
128
+ rfd4-generate inputs=data/benchmarks/monomer/monomer-short.cif \
129
+ hf_ckpt_path=nvidia/RFD4-Proteina \
130
+ hf_ckpt_revision=a06e2509c7033b010fc43614e9fd840303588502 \
131
+ num_replicates=1 out_dir=./runs/first '~metrics' compile_lengths=null
132
+ ```
133
+
134
+ Find the generated CIF under `runs/first/inference_samples/`. This first run disables external scoring and compilation to keep setup simple. For repeated runs, use the normal compiled configuration described in [running the model](https://rfd4-proteina.org/latest/how-to/run-the-model.html); compilation has an initial warmup cost.
135
+
136
+ ### 3. Design a PD-L1 binder or an enzyme scaffold
137
+
138
+ ```bash
139
+ hf download nvidia/RFD4-Proteina \
140
+ --revision a06e2509c7033b010fc43614e9fd840303588502 \
141
+ --local-dir ./rfd4-examples \
142
+ examples/protein_binder/input.cif examples/enzyme_design/input.cif
143
+
144
+ rfd4-generate inputs=./rfd4-examples/examples/protein_binder/input.cif \
145
+ hf_ckpt_path=nvidia/RFD4-Proteina \
146
+ hf_ckpt_revision=a06e2509c7033b010fc43614e9fd840303588502 \
147
+ num_replicates=1 out_dir=./runs/pdl1 '~metrics' compile_lengths=null
148
+ ```
149
+
150
+ For enzyme motif scaffolding, replace the input with `./rfd4-examples/examples/enzyme_design/input.cif` and choose a new output directory. The annotations specify what to hold fixed and what to design; an ordinary, unannotated structure is not a complete design specification.
151
+
152
+ | Design goal | Start here |
153
+ |---|---|
154
+ | Protein–protein binder | [PD-L1 input](examples/protein_binder/input.cif) · [example output](examples/protein_binder/design.cif) |
155
+ | Enzyme motif scaffolding | [Enzyme input](examples/enzyme_design/input.cif) · [example output](examples/enzyme_design/design.cif) |
156
+ | Antibody design | [HER2 input](examples/antibody_design/input.cif) · [example output](examples/antibody_design/design.cif) |
157
+ | Unconditional monomer | [100-residue spec](examples/monomer.cif) — a length specification, no input coordinates |
158
+ | Prepare your own conditions | [Annotated CIFs and Python input preparation](https://rfd4-proteina.org/latest/how-to/input_prep/index.html) · [YAML specifications](https://rfd4-proteina.org/latest/how-to/input_prep/yaml-json.html) |
159
+ | Interactive design | [AtomWorks Studio](https://huggingface.co/spaces/nvidia/rfd4-proteina-space) on a dedicated GPU pool · [the same Studio on ZeroGPU](https://huggingface.co/spaces/nvidia/rfd4-proteina-zerogpu) if you would rather spend your own Hugging Face quota. Both screen every input and design with SecureDNA and refuse flagged requests. |
160
+
161
+ ## Loading the model
162
+
163
+ ### Native RFD4 (this repository)
164
+
165
+ Load the matched model and autoencoder once and reuse them across runs:
166
+
167
+ ```python
168
+ from pathlib import Path
169
+ from huggingface_hub import hf_hub_download
170
+ from hydra import compose, initialize_config_dir
171
+ import rootutils
172
+ import rfproteina
173
+ from rfproteina.proteina import Proteina
174
+ from rfproteina.generate import run_generation
175
+
176
+ revision = "a06e2509c7033b010fc43614e9fd840303588502"
177
+ input_cif = hf_hub_download(
178
+ "nvidia/RFD4-Proteina", "examples/enzyme_design/input.cif", revision=revision,
179
+ )
180
+ model = Proteina.from_pretrained("nvidia/RFD4-Proteina", revision=revision)
181
+ config_dir = rootutils.find_root(rfproteina.__file__, indicator=".project-root") / "configs"
182
+ with initialize_config_dir(config_dir=str(config_dir), version_base="1.3"):
183
+ cfg = compose(config_name="inference", overrides=[
184
+ f"inputs={input_cif}", "out_dir=./runs/enzyme",
185
+ "num_replicates=1", "~metrics", "compile_lengths=null",
186
+ ])
187
+ run_generation(cfg, model=model)
188
+ print(list(Path("runs/enzyme/inference_samples").rglob("*_generated.cif*")))
189
+ ```
190
+
191
+ `from_pretrained` loads in evaluation mode on CPU by default; `run_generation` uses the native CUDA inference workflow. A local exported bundle directory also works as the model argument.
192
+
193
+ ### Diffusers
194
+
195
+ ```bash
196
+ # RFD4-Proteina was merged into Diffusers on 2026-10-20 and ships in the next release.
197
+ # Until that release, install Diffusers from git:
198
+ pip install git+https://github.com/huggingface/diffusers git+https://huggingface.co/nvidia/RFD4-Proteina-prep@c9959efdebd75ac7fe9129fbbf0e291a6e0487d5 accelerate
199
+
200
+ # Once the release is out, this is enough:
201
+ # pip install diffusers git+https://huggingface.co/nvidia/RFD4-Proteina-prep@c9959efdebd75ac7fe9129fbbf0e291a6e0487d5
202
+ ```
203
+
204
+ ```python
205
+ import torch
206
+ from diffusers import RFD4ProteinaPipeline
207
+ from rfd4_proteina_prep import prepare_cif, write_cif
208
+
209
+ pipe = RFD4ProteinaPipeline.from_pretrained("nvidia/RFD4-Proteina-Diffusers", dtype=torch.bfloat16).to("cuda:0")
210
+ features, atoms = prepare_cif("input.cif", seed=42, device=pipe.device)
211
+ required = ["padding_mask", "is_in_context_conditioning", "supports_local_latent", "token_type",
212
+ "atom37_mask", "coords_nm"]
213
+ conditioning = {name: value for name, value in features.items() if name not in required}
214
+ output = pipe(
215
+ padding_mask=features["padding_mask"],
216
+ is_in_context_conditioning=features["is_in_context_conditioning"],
217
+ supports_local_latent=features["supports_local_latent"],
218
+ token_type=features["token_type"],
219
+ atom37_mask=features["atom37_mask"],
220
+ conditioning_features=conditioning,
221
+ generator=torch.Generator("cuda:0").manual_seed(42),
222
+ )
223
+ write_cif(output, features, atoms, "design.cif")
224
+ ```
225
+
226
+ This needs Diffusers installed from git, as shown above. Partial diffusion and inverse folding work in this port. See the [repository](https://huggingface.co/nvidia/RFD4-Proteina-Diffusers) for details.
227
+
228
+ ### Compute presets
229
+
230
+ The released recipes differ only in how often the trunk is recycled per sampling step, as `(steps, recycle depth)` bands over 60 steps. `thinking_low|medium|high` are the front-loaded schedules, and the default is medium. The native runtime also has flat and mid-peak schedules for each budget; see the [running guide](https://rfd4-proteina.org/latest/how-to/run-the-model.html#recycling-schedules) for which one suits a task.
231
+
232
+ | Preset | Bands | Trunk passes |
233
+ |---|---|---:|
234
+ | Low | `[[10,4],[10,3],[10,2],[30,1]]` | 120 |
235
+ | **Medium (default)** | `[[5,6],[10,5],[5,4],[10,3],[20,2],[10,1]]` | **180** |
236
+ | High | `[[10,9],[10,8],[10,5],[20,3],[10,2]]` | 300 |
237
+
238
+ In the native runtime select one with `recycle_strategy=thinking_low|thinking_medium|thinking_high`.
239
+
240
+ ## Hardware requirements
241
+
242
+ Tested and confirmed on the following NVIDIA GPUs. Linux x86_64, Python 3.12, and a CUDA 13-compatible driver are required.
243
+
244
+ | Architecture | Data center | Workstation and desktop |
245
+ |---|---|---|
246
+ | Blackwell | B200, B300, GB200, GB300, RTX PRO 6000 Blackwell Server Edition | RTX PRO 6000 Blackwell Workstation Edition, GB10 (DGX Spark) |
247
+ | Hopper | H100 SXM, H100 PCIe, H100 NVL, H200, H200 NVL, GH200 144GB | |
248
+ | Ada Lovelace | L40S | RTX 6000 Ada Generation |
249
+ | Ampere | A100 SXM4 40GB and 80GB, A100 PCIe 40GB and 80GB | |
250
+
251
+ Weights are 1.74 GB in FP32. Peak GPU memory depends on token count, batch size and recycle depth rather than on the weights; the quickstart above runs at batch size 1 with compilation disabled. Refolding and metrics need the selected oracle's own environment and weights, which are a separate download.
252
+
253
+ ## Measured design efficiency
254
+
255
+ > **These numbers were measured on the previous release pair**, in the October 9, 2026 frozen project benchmark, which predates the current checkpoints. They have not been re-measured on the current release and should not be read as describing it.
256
+
257
+ The table uses one consistent high-thinking setting for RFD4 and the strongest non-RFD4 comparator in each family. **Metric:** mean per-target structurally distinct computational passers per GPU-hour, including generation and common AF3 evaluation. Higher is better.
258
+
259
+ | Benchmark | Targets | RFD4-Proteína, high thinking | Best other method in this comparison |
260
+ |---|---:|---:|---|
261
+ | Protein binders | 24 | **71.60** | Proteina-Complexa · 13.24 |
262
+ | Small-molecule binders | 77 | **112.25** | BoltzGen · 15.48 |
263
+ | Atomic Motif Enzyme | 37 | **91.18** | AlphaProtein-novo (Codesign) · 26.02 |
264
+ | DNA binders | 3 | **41.97** | BoltzGen (fixed DNA) · 9.46 |
265
+
266
+ There are 200 original attempts per target. Failed attempts consume time; setup, compilation, CPU scoring, and idle allocation time are excluded. Values are means of target-level rates, not pooled totals or single-design latency.
267
+
268
+ Conditioning differs between some methods: PPI uses RFD4 backbone conditioning versus all-atom conditioning for several baselines; DNA compares unfixed RFD4 inputs with fixed-DNA baselines; Proteina-Complexa's small-molecule route uses fixed ligands. Only three DNA targets are included. Computational passers do not establish experimental activity.
269
+
270
+ [Benchmark provenance](benchmarks/provenance.json) · [Full comparison table](benchmarks/summary.csv) · [Per-target source data](benchmarks/source_data.csv) · [Benchmark methods and pass criteria](benchmarks/METHODS.md)
271
+
272
+ ## Training data and limitations
273
+
274
+ The companion [dataset card](https://huggingface.co/datasets/nvidia/RFD4-Proteina) describes the available training snapshot, component-specific filters, source terms, and loading format. Its upload date or proposed sampling mix should not be interpreted as the exact training history of every checkpoint.
275
+
276
+ Use the model for computational biomolecular design and follow designs with appropriate structural and experimental evaluation. Performance depends on the target, input annotations, sampling settings, and training coverage. Generation-only quickstarts omit external refolding and scoring; see [evaluation and metrics](https://rfd4-proteina.org/latest/how-to/evaluate-with-metrics.html) to add them.
277
+
278
+ <details>
279
+ <summary><strong>Other repository files</strong></summary>
280
+
281
+ **Third-party software and weights.** The BSD 3-Clause license covers the RFD4-Proteína model, weights and code only. The optional `alphafold2/runtime/` and `openfold3/runtime/` container images package third-party software (AlphaFold2- and AlphaFold3-derived code, OpenFold3 and their dependencies) that stays under its own upstream licenses, some of which restrict commercial use or redistribution. The ported OpenFold3 weights in `openfold3/weights/` are Apache-2.0 (see the provenance file). Review each upstream license before using the images, and do not assume the BSD license applies to them. Neither oracle is needed for generation.
282
+
283
+ Optional [AF3/OpenFold3 runtime images](openfold3/runtime) are available for Linux x86_64 with CUDA 12 or CUDA 13 in Apptainer (`.sif`), SquashFS (`.sqsh`) and Docker (`.docker.tar`) formats, with SHA-256 checksums. [OpenFold3 weights](openfold3/weights) are downloaded separately. These files are only needed for refolding and metrics; generation does not use them.
284
+
285
+ </details>
286
+
287
+ ## License
288
+
289
+ The RFD4-Proteína model and weights are released under the [BSD 3-Clause License](LICENSE), Copyright (c) 2026, Institute for Protein Design, University of Washington. Optional third-party assets distributed alongside the model — the AlphaFold2 and OpenFold3/AlphaFold3 refolding images and the OpenFold3 payload — carry their own upstream terms; see the note under *Other repository files* and their provenance files. The training data has its own source terms, described on the [dataset card](https://huggingface.co/datasets/nvidia/RFD4-Proteina).
290
+
291
+ ## Citation
292
+
293
+ When reporting results, cite the RFD4-Proteína project together with the exact model revision, code revision, seed, inputs and sampling configuration you used, plus any source datasets and evaluation tools.
294
+
295
+ ```bibtex
296
+ @software{rfd4_proteina_2026,
297
+ title = {RFD4-Proteína: joint sequence and all-atom structure design},
298
+ author = {{Baker Lab, Institute for Protein Design, University of Washington} and {NVIDIA GenAIR}},
299
+ year = {2026},
300
+ url = {https://huggingface.co/nvidia/RFD4-Proteina},
301
+ note = {Model revision a06e2509c7033b010fc43614e9fd840303588502}
302
+ }
303
+ ```
304
+
305
+ ## Attribution and support
306
+
307
+ Developed by the Baker Lab at the Institute for Protein Design, University of Washington, and NVIDIA GenAIR.
308
+
309
+ [Code and issues](https://github.com/RosettaCommons/RFD4-Proteina/issues) · [Model history](https://huggingface.co/nvidia/RFD4-Proteina/commits/main) · [License](LICENSE)
adapter/adapter_config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": null,
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": false,
14
+ "init_lora_weights": true,
15
+ "kasa_config": null,
16
+ "layer_replication": null,
17
+ "layers_pattern": null,
18
+ "layers_to_transform": null,
19
+ "loftq_config": {},
20
+ "lora_alpha": 64.0,
21
+ "lora_bias": false,
22
+ "lora_dropout": 0.0,
23
+ "lora_ga_config": null,
24
+ "megatron_config": null,
25
+ "megatron_core": "megatron.core",
26
+ "modules_to_save": null,
27
+ "monteclora_config": null,
28
+ "peft_type": "LORA",
29
+ "peft_version": "0.21.2",
30
+ "qalora_group_size": 16,
31
+ "r": 32,
32
+ "rank_pattern": {},
33
+ "revision": null,
34
+ "target_modules": [
35
+ "to_qkv",
36
+ "to_out_node"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": null,
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false,
45
+ "velora_config": null
46
+ }
adapter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b0762a73b543dfd88958267ba93a729fd91b31f2e18dba2d34c302260f50402
3
+ size 8264360
ae_checkpoints/rfd4_proteina_ae-ema.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24536372f3fc9a5083b6b813baa80630d2287a5d61fbf2f126c9735802e1d21a
3
+ size 1029340246
ae_checkpoints/rfd4_proteina_ae-ema.ckpt.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 24536372f3fc9a5083b6b813baa80630d2287a5d61fbf2f126c9735802e1d21a rfd4_proteina_ae-ema.ckpt
alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sif ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62b6531b87bafbbfb584ae2c341b9bab4591a733ffbc39abd2d6dd4b7ac214c6
3
+ size 16566288384
alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sif.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 62b6531b87bafbbfb584ae2c341b9bab4591a733ffbc39abd2d6dd4b7ac214c6 alphafold2-pallas-cuda12-amd64.sif
alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sqsh ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b037a78501b5d3c2d39ebdf9faecb3b4e0e7d7adc7bf4693f33699ba2b7ddc9
3
+ size 16566255616
alphafold2/runtime/alphafold2-pallas-cuda12-amd64.sqsh.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 8b037a78501b5d3c2d39ebdf9faecb3b4e0e7d7adc7bf4693f33699ba2b7ddc9 alphafold2-pallas-cuda12-amd64.sqsh
alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sif ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5160d179a0f11608e176f8883c8e9c644ab5c6a6b92bf05c2ca5b7a5ecb8d3f
3
+ size 6051827712
alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sif.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ b5160d179a0f11608e176f8883c8e9c644ab5c6a6b92bf05c2ca5b7a5ecb8d3f alphafold2-pallas-cuda13-amd64.sif
alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sqsh ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9982da8f06ee257366e31f395c0d006bb28ada4d22137e5561b5b0f908c9804a
3
+ size 12401594368
alphafold2/runtime/alphafold2-pallas-cuda13-amd64.sqsh.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 9982da8f06ee257366e31f395c0d006bb28ada4d22137e5561b5b0f908c9804a alphafold2-pallas-cuda13-amd64.sqsh
assets/design-gallery.png ADDED

Git LFS Details

  • SHA256: 037564293ed2582451861616539f35d9cfa6ee94e6ba5698d8df95df2f9cf02a
  • Pointer size: 132 Bytes
  • Size of remote file: 1.28 MB
assets/examples/antibody_design.png ADDED

Git LFS Details

  • SHA256: 6f00966ca05c0ed716e128820f0a9c2581ace24a32855d58b1875551bad65500
  • Pointer size: 131 Bytes
  • Size of remote file: 211 kB
assets/examples/enzyme_design.png ADDED

Git LFS Details

  • SHA256: db631416823bd2a0419d1d8902950cfad25a2ab230e13cf29517bdfce1c25939
  • Pointer size: 131 Bytes
  • Size of remote file: 147 kB
assets/examples/protein_binder.png ADDED

Git LFS Details

  • SHA256: 902488410d3cdf556f1f6b3aa1f06b49ab0b2af32c065ba0b187221b07ffbcfa
  • Pointer size: 131 Bytes
  • Size of remote file: 212 kB
assets/wordmark-dark.webp ADDED
assets/wordmark.webp ADDED
base/config.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "rfd4_proteina",
3
+ "format_version": 1,
4
+ "weights_sha256": "10e90a76e3b09c94a700d2f67e89ed660af7f3018e3f2209b5d52f24d00fbd39",
5
+ "config": {
6
+ "nn": {
7
+ "name": "local_latents_transformer",
8
+ "output_parameterization": {
9
+ "bb_ca": "v",
10
+ "local_latents": "v"
11
+ },
12
+ "sequence_readout": {
13
+ "enabled": true,
14
+ "detach": true,
15
+ "num_classes": 147
16
+ },
17
+ "nlayers": 14,
18
+ "token_dim": 768,
19
+ "nheads": 12,
20
+ "parallel_mha_transition": false,
21
+ "token_type_emb_dim": 32,
22
+ "cath_emb_dim": 32,
23
+ "strict_feats": false,
24
+ "noise_align_chirality": true,
25
+ "noise_align_internal_coords": false,
26
+ "noise_align_chirality_translate": false,
27
+ "noise_align_internal_coords_translate": false,
28
+ "use_atom37_factory": true,
29
+ "feats_seq": [
30
+ "xt_bb_ca",
31
+ "xt_local_latents",
32
+ "token_type",
33
+ "sequence_probabilities",
34
+ "cond_a37coors_nm",
35
+ "cond_a37coors_internal_nm",
36
+ "rasa_binned",
37
+ "secondary_structure",
38
+ "hbond_modernized",
39
+ "pistack",
40
+ "pication",
41
+ "saltbridge",
42
+ "hydrophobic",
43
+ "metal_coord",
44
+ "na_ss_state",
45
+ "hotspot",
46
+ "atom_charge",
47
+ "atomic_element",
48
+ "coord_group_seq",
49
+ "chirality",
50
+ "chain_centroid_xyz"
51
+ ],
52
+ "feats_cond_seq": [
53
+ "time_emb_bb_ca",
54
+ "time_emb_local_latents",
55
+ "cath_code",
56
+ "recycle_remaining"
57
+ ],
58
+ "dim_cond": 256,
59
+ "idx_emb_dim": 256,
60
+ "t_emb_dim": 256,
61
+ "feats_pair_repr": [
62
+ "xt_bb_ca_pair_dists",
63
+ "bond_mask",
64
+ "bond_order",
65
+ "is_aromatic",
66
+ "index_group_pair",
67
+ "molecule_idx_pair",
68
+ "molecule_entity_pair",
69
+ "token_bond_path_length",
70
+ "canonical_polymer_ordering",
71
+ "na_ss_pair"
72
+ ],
73
+ "feats_pair_cond": [
74
+ "time_emb_bb_ca",
75
+ "time_emb_local_latents"
76
+ ],
77
+ "xt_pair_dist_dim": 30,
78
+ "xt_pair_dist_min": 0.1,
79
+ "xt_pair_dist_max": 3,
80
+ "seq_sep_dim": 64,
81
+ "pair_repr_dim": 384,
82
+ "update_pair_repr": true,
83
+ "update_pair_repr_every_n": 2,
84
+ "use_tri_mult": true,
85
+ "tri_mult_c": 256,
86
+ "tri_mult_bias": false,
87
+ "tri_mult_backend": "cuequivariance",
88
+ "recycle": {
89
+ "enabled": true,
90
+ "grad": "last",
91
+ "training_max_recycles": 9,
92
+ "condition_probability": 0.5
93
+ },
94
+ "checkpointing": "none",
95
+ "pair_write": "hadamard",
96
+ "use_qkln": true,
97
+ "attention_type": "torch",
98
+ "use_torch_compile": false,
99
+ "norm_type": "rmsnorm_torch_fused"
100
+ },
101
+ "product_flowmatcher": {
102
+ "bb_ca": {
103
+ "zero_com_noise": true,
104
+ "guidance_enabled": true,
105
+ "dim": 3,
106
+ "stochastic_centering_mode": "x_0",
107
+ "stochastic_centering_scale": 0.2,
108
+ "stochastic_centering_scope": "per_group",
109
+ "stochastic_centering_group_keys": [
110
+ "molecule_iid"
111
+ ]
112
+ },
113
+ "local_latents": {
114
+ "zero_com_noise": false,
115
+ "guidance_enabled": true,
116
+ "dim": 32,
117
+ "stochastic_centering_mode": "none",
118
+ "stochastic_centering_scale": null
119
+ }
120
+ },
121
+ "loss": {
122
+ "signed_volume_weight": 2.5,
123
+ "sequence_prediction": {
124
+ "enabled": true,
125
+ "t_min": 0.0,
126
+ "t_max": 1.0
127
+ },
128
+ "t_distribution": {
129
+ "bb_ca": {
130
+ "name": "mix_unif_beta",
131
+ "p1": 1.9,
132
+ "p2": 1.0,
133
+ "p3": 0.02
134
+ },
135
+ "local_latents": {
136
+ "name": "mix_unif_beta",
137
+ "p1": 1.0,
138
+ "p2": 1.5,
139
+ "p3": 0.2
140
+ },
141
+ "shared_groups": null
142
+ }
143
+ },
144
+ "training": {
145
+ "exploration_k": 4,
146
+ "exploration_losses": [
147
+ "bb_ca",
148
+ "local_latents"
149
+ ],
150
+ "n_recycle_max": 9,
151
+ "n_recycle_sampler": "poisson",
152
+ "n_recycle_lambda": 3.0,
153
+ "n_recycle_sync": true,
154
+ "n_recycle_val_loss": 3,
155
+ "motif_conditioning": false
156
+ },
157
+ "opt": {
158
+ "lr": 0.0001,
159
+ "muon_warmup_steps": 0,
160
+ "adam_warmup_steps": 10000,
161
+ "gradient_clip_val": 1.0,
162
+ "max_epochs": 10000000,
163
+ "accumulate_grad_batches": 1,
164
+ "dist_strategy": "ddp",
165
+ "steps_per_epoch": 2000,
166
+ "val_every_n_epochs": 10,
167
+ "val_start_epoch": 0,
168
+ "compute_val_loss": false,
169
+ "foldseek_timeout_minutes": 25,
170
+ "muon_exclude_param_patterns": [
171
+ "_factory\\.",
172
+ "^nn\\.(local_latents_linear|ca_linear|insertion_q_proj|insertion_k_proj)\\.",
173
+ "^nn\\.sequence_readout\\."
174
+ ],
175
+ "muon_adjust_lr_fn": "match_rms_adamw",
176
+ "muon_grouped": true,
177
+ "num_sanity_val_steps": 2,
178
+ "coordinate_descent_tuning": false
179
+ },
180
+ "autoencoder_config": {
181
+ "nn_ae": {
182
+ "name": "transformer_ae_aa_rna_dna_glycan_ncaa",
183
+ "latent_z_dim": 32,
184
+ "allow_oxt": true,
185
+ "use_torch_compile": false,
186
+ "torch_compile_fullgraph": false,
187
+ "n_residue_types": 147,
188
+ "encoder": {
189
+ "nlayers": 12,
190
+ "token_dim": 768,
191
+ "nheads": 12,
192
+ "parallel_mha_transition": false,
193
+ "token_type_emb_dim": 32,
194
+ "strict_feats": false,
195
+ "feats_seq": [
196
+ "token_type",
197
+ "x1_a37coors_nm",
198
+ "x1_a37coors_nm_rel",
199
+ "chain_idx_seq",
200
+ "incident_bond_type_counts",
201
+ "atom37_element_charge_embedding"
202
+ ],
203
+ "feats_cond_seq": null,
204
+ "dim_cond": 128,
205
+ "idx_emb_dim": 128,
206
+ "feats_pair_repr": [
207
+ "token_bond_path_length",
208
+ "canonical_polymer_ordering",
209
+ "molecule_idx_pair",
210
+ "x1_bb_pair_dists_nm"
211
+ ],
212
+ "seq_sep_dim": 64,
213
+ "pair_repr_dim": 256,
214
+ "update_pair_repr": false,
215
+ "update_pair_repr_every_n": 3,
216
+ "use_tri_mult": false,
217
+ "use_attn_pair_bias": false,
218
+ "use_qkln": true,
219
+ "normalize_latent": false,
220
+ "latent_z_dim": 32,
221
+ "n_residue_types": 147,
222
+ "charge_emb_dim": 16
223
+ },
224
+ "decoder": {
225
+ "nlayers": 12,
226
+ "token_dim": 768,
227
+ "abs_coors": false,
228
+ "nheads": 12,
229
+ "token_type_emb_dim": 32,
230
+ "parallel_mha_transition": false,
231
+ "strict_feats": false,
232
+ "feats_seq": [
233
+ "ca_coors_nm",
234
+ "z_latent_seq",
235
+ "token_type",
236
+ "cond_residue_type"
237
+ ],
238
+ "feats_cond_seq": null,
239
+ "dim_cond": 128,
240
+ "idx_emb_dim": 128,
241
+ "feats_pair_repr": [
242
+ "token_bond_path_length",
243
+ "canonical_polymer_ordering",
244
+ "molecule_idx_pair",
245
+ "ca_coors_nm_pair_dists"
246
+ ],
247
+ "seq_sep_dim": 64,
248
+ "pair_repr_dim": 256,
249
+ "update_pair_repr": false,
250
+ "update_pair_repr_every_n": 3,
251
+ "use_tri_mult": false,
252
+ "use_attn_pair_bias": false,
253
+ "use_qkln": true,
254
+ "latent_z_dim": 32,
255
+ "n_residue_types": 147
256
+ }
257
+ }
258
+ },
259
+ "path_length_search_cutoffs": [
260
+ 64
261
+ ]
262
+ }
263
+ }
base/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10e90a76e3b09c94a700d2f67e89ed660af7f3018e3f2209b5d52f24d00fbd39
3
+ size 1742660596
benchmarks/METHODS.md ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Supplement methods
2
+
3
+ This release uses the frozen cohorts in `evidence/freeze.json`, not every run in the live catalog. It includes current Figure 2 methods and the additional RFD4 High conditioning arms. Antibody, HalluDesign, full-pipeline BindCraft and Genie3 search are excluded. The per-run evidence is under `models/<model>/provenance/`; exact native inputs and timing receipts take precedence over any current upstream default.
4
+
5
+ ## Common accounting and timing
6
+
7
+ Each model × protocol × target has **200 original attempts**, including native failures and unmatched required motifs. We do not replace rejected originals or select outputs with the common AF3 evaluation. Protein-Hunter and RFD4 search have explicit intrinsic selection; those stages count toward their generation cost. Novo's two arms share original scaffolds and are not independent samples.
8
+
9
+ `generation_gpu_s` is the complete stored warm generation cost, including required sequence design, intrinsic folding/search, and native gaps inside that timed boundary. It excludes setup, loading, warmup and compilation where separately measured. It is not a diffusion-only timer. Batched costs are amortized across the batch's originals; batch shares are not individually measured latency. `timing_workers.csv` retains available setup and warm totals. `batch_size` is the measured actual batch, including tails—not a claim of maximal possible throughput. Unknown metadata remains blank. Exact-length/bucket compilation limitations are retained in executed receipts; the release does not assert that all models were equally optimized.
10
+
11
+ `af3_gpu_s` is separate common evaluation time. Default efficiency is the **arithmetic mean across targets of successes / (generation + AF3 GPU time)**, multiplied by 600 for per-10-GPU-minute units. Failed originals and their consumed time stay included. MPNN and native-fold columns are components already included in generation: **do not add them again**. A zero means the component did not run or used no GPU; blank means it was not separately measured. A missing AF3 cost excludes that configuration from an inclusive efficiency comparison, not its successes from rate comparisons.
12
+
13
+ Timing exceptions: ODesign retains the user-approved original-generation timing proxy, which includes its old MPNN step rather than separately adding corrected MPNN. Complexa fixed-small-molecule costs are registered batched remeasurements, not original batch-1 wall time. Historical RFD4 PPI all-atom uses the approved backbone timing proxy and has no measured AF3 cost. These are identified in `timing_basis`; they are not disguised as direct per-design measurements.
14
+
15
+ ## Model recipes and adaptations
16
+
17
+ | Model | Executed generation and input/output considerations | Author reference |
18
+ |---|---|---|
19
+ | RFD4-Proteina | Joint sequence/structure generation with stored Low/Medium/High schedules; exact schedule/checkpoint comes from each native config. Search is a separate intrinsic workflow: AF2-Multimer for PPI and OF3 for the other benchmarks, with all search work timed. Preserve native motif correspondence; unmatched required motifs fail. Historical PPI-AA validation is explicitly separate from the newer BB campaign. | [RFD4 repository](https://github.com/RosettaCommons/RFD4-Proteina) |
20
+ | RFdiffusion3 | Native 200 steps, CFG off, step size 1.5; one LigandMPNN sequence at temperature 0.1, no downstream pipeline screening. DNA uses the DNA-finetuned checkpoint. Native JSON/derivatives preserve shared lengths and correspondence. Approved PPI OXT omission and native M0110 bond/charge/preprocessing losses are retained. Ligands can follow native reference conformers; no stereo repair. | [Foundry](https://github.com/RosettaCommons/foundry) |
21
+ | BoltzGen | Design: 500 steps/3 recycles; inverse folding: 200 steps/3 recycles, one final sequence. Original indices 0–99 use diverse, 100–199 adherence, preserving the native quarter settings across shards. No terminal folding/filtering/ranking. Atom hotspots become residue hotspots; PPI is all-atom. Seven heme cases use disclosed native bond-type projections; difficult chemistry uses source-derived archives, never author benchmark molecules. | [BoltzGen](https://github.com/HannesStark/boltzgen) |
22
+ | Proteina-Complexa | One 400-step joint draw, no additional MPNN/selection. PPI uses V1; AME V2/LoRA32; fixed small molecules use the corrected dedicated V1/LoRA32 ligand route. Native matcher is retained: unmatched motifs fail downstream. PPI and ligand inputs use native atom37/template features and residue-projected hotspots. No DNA or fully unfixed ligand route. | [Proteina-Complexa](https://github.com/NVIDIA-BioNeMo/Proteina-Complexa) |
23
+ | DISCO | Native diverse/max joint draw, 200 diffusion steps/four recycles. No extra sequence designer. SDF/JSON derives from shared CIF chemistry, never checked-in benchmark SDFs. Same-target/exact-length batches; all batch members retained. Native hydrogen/bond-order losses are documented, not repaired. | [DISCO](https://github.com/DISCO-design/DISCO) |
24
+ | Genie3 | One 100-step generative draw and one CA-MPNN sequence at 0.1. **No search or ColabFold.** Force the original hotspot, using target CA and source hotspot sidechain geometry, not extended contacts. Native CA-only outputs stay CA-only and are evaluated with binder CA RMSD. Some executed input receipts reference an earlier source release; exact executed files remain preserved, not silently rebound to a newer queue. | [Genie3](https://github.com/aqlaboratory/genie3) |
25
+ | BindCraft (single backbone) | One 75/45/5/15 optimization trajectory, then one soluble ProteinMPNN v_48_020 sequence at 0.1. Three continuation gates bypassed to retain bad trajectories. No post-MPNN fold/relax/filter/rank. Native independent batch 1; exact lengths warmed. Native interface-fixed residues retained. Output sidechains incompatible with the new sequence become unobserved rather than being grafted. | [BindCraft tutorial](https://github.com/martinpacesa/BindCraft/wiki/De-novo-binder-design-with-BindCraft) |
26
+ | BindCraft2 (single backbone) | One 50/25/45/5/15 trajectory, then one native soluble-MPNN redesign at 0.1. Continuation gates disabled; no independent validation or relaxation. Native independent batch 1, unchunked; resident AF2/MPNN pair, separately warmed signatures. Preserve exact native redesign constraints and report omitted incompatible sidechains. | [BindCraft2](https://github.com/PacesaLab/BindCraft2) |
27
+ | Protein-Hunter | Seven approved native refinement cycles, 100%-X initialization, 200 steps/3 recycles, MPNN 0.1. SolubleMPNN for PPI; LigandMPNN for ligands/DNA. Retain highest native ipTM cycle among those with ≤20% alanine. All intrinsic cycles are timed. Native rejection and writer/chemistry defects are not rescued. PPI guidance is residue hotspots and CA/CB/frame guidance, not hard sidechain enforcement. | [Protein-Hunter](https://github.com/yehlincho/Protein-Hunter) |
28
+ | ESMFold2 inversion | 150 inversion steps, two pinned models/four intrinsic hero critics. Critic 0 is predesignated; no common-AF3 ranking. Target sequence/residue hotspots, not target coordinates. Multichain extension preserves all target copies and native whole-complex ipTM, including target–target contributions. Intrinsic critics are not separate benchmark attempts. | [ESM binder-design tutorial](https://github.com/Biohub/esm/blob/main/cookbook/tutorials/binder_design.ipynb) |
29
+ | RFdiffusion | 50-step Complex_base draw with example CA/frame noise scales 0; one ProteinMPNN v_48_020 sequence at 0.1. Resident native batch 1. No Rosetta/refold/selection. Explicit contig/chain maps retain target copies and gaps. Preserve native backbone-only output and atom omissions. | [RFdiffusion binder design](https://github.com/RosettaCommons/RFdiffusion#binder-design) |
30
+ | RFdiffusion2 | AME RFD_140 checkpoint, 100 native flow steps, one LigandMPNN 0.1 sequence with native packing/context. Resident native batch 1. M0110 uses the explicitly approved disconnected ligand projection: preserve source atoms/coordinates but omit two bonds in the native derivative. This is not equivalent physical topology. Native guidepost matching/idealization is retained. | [RFdiffusion2](https://github.com/RosettaCommons/RFdiffusion2) |
31
+ | ODesign | Native scaffold generation followed by the corrected one-LigandMPNN0.1, source-mask-fixed sequence route. No terminal AF3 selection. Native AME receives disclosed extra observed motif context and is not a strict tip-only baseline. Fixed-ligand and fixed-DNA membership is labeled explicitly. Original-generation timing is a proxy, as described above. | [ODesign](https://github.com/OTeam-AI4S/ODesign) |
32
+ | AlphaProtein-novo | Native 1000-step joint codesign. Paired LigandMPNN arm uses one sequence at 0.1, checkpoint 030, atom context on, sidechain context off, no packing. CPU MPNN adds zero GPU seconds; shared scaffold generation time is inherited, not omitted. Native APN1 composite-ligand handling of M0110 changes interface interpretation and is disclosed. | [AlphaProtein-novo](https://github.com/google-deepmind/alphaprotein-novo) |
33
+
34
+ Native source/current workflow snapshots and immutable repository citations are listed in `references.csv`. They explain the implementation; executed receipts—not the newest README—identify what each cohort actually ran.
35
+
36
+ ## Common AF3 scoring
37
+
38
+ The existing accelerated external-eval AF3 outputs are reused; this release launches no generation, folding or rescoring. Protein targets are templated from propagated coordinate-conditioning information using stored coordinates. DNA alignment includes the target DNA in both conditioning modes. Template sanity metrics are diagnostic, not substitutions for binder accuracy. Native biological-symmetry handling is retained in stored scores; do not derive alternative scores merely by relabeling chains.
39
+
40
+ | Benchmark | Current pass criterion |
41
+ |---|---|
42
+ | PPI, 24 targets | AF3 binder pTM ≥80 on the native 0–100 scale; minimum interface PAE ≤1.5 Å; target-backbone-aligned binder backbone RMSD ≤2.5 Å (CA-only for Genie3). Required target mapping/conditioning must be valid. Target mutations at a hotspot or within 5 Å of binder heavy atoms in design or refold fail; distal substitutions may pass. |
43
+ | Small molecule, 77 targets | AF3 minimum interface PAE ≤1.5 Å, chain-of-interest ipTM ≥80/100, protein-backbone RMSD ≤1.5 Å, and protein-backbone-aligned ligand RMSD ≤5 Å. This is `sm_binder`, not the looser historical `disco` criterion. |
44
+ | Original DISCO, same 77 targets | Protein-backbone RMSD ≤2 Å and maximum ligand-centroid displacement ≤2 Å after protein-backbone alignment. No AF3 PAE/ipTM cutoff. Stored verdicts and independently computed passing-cluster labels are retained separately from `sm_binder`. The criteria are not mathematically nested because the ligand-position tests differ. |
45
+ | Atomic Motif Enzyme (AME), 37 targets | Protein-backbone RMSD ≤2 Å, required motif all-atom RMSD ≤1.5 Å, and AF3 minimum interface PAE **<1.5 Å**. Motif RMSD aligns/scores mapped conditioned protein atoms against stored input references, respecting coordinate groups. Native-unmatched/sequence-failed motifs fail; missing metrics are not assigned zero. Clashes are diagnostic only in this current policy. |
46
+ | DNA, 3 targets | AF3 minimum interface PAE ≤1.5 Å and DNA-target-aligned binder backbone RMSD ≤8 Å. Alignment uses all target DNA atoms in both fixed/unfixed modes. |
47
+
48
+ Ordinary rate = passing originals / 200. Clustered rate = distinct clusters among passing originals / 200. Clusters are computed **within each model/protocol/target**, using generated binder/protein structures and Foldseek TM threshold 0.5, alignment type 1, minimum sequence identity 0; they are not pooled between targets or models. Dark bar plus light extension is the ordinary rate; dark alone is clustered. These are algorithmic structural clusters, not experimental successes or globally distinct proteins.
49
+
50
+ ## Supplement selection and conditioning comparisons
51
+
52
+ Per-target panels include RFD4 High, the best other generative model, and best non-generative model by mean clustered rate (ordinary rate breaks ties). The requested AME panel instead shows four: RFD4 High, AlphaProtein-novo Codesign, Proteina-Complexa and RFdiffusion3. Both small-molecule criteria get separate panels and comparator selection. Full selection and all model values remain in CSVs.
53
+
54
+ The High conditioning panels use all 200 originals per target and do not assume the desired direction. **PPI-AA and PPI-BB are different campaigns**; other settings/checkpoints can confound conditioning. AME compares *partially* to fully fixed ligand, not fully unfixed. Three NAD small-molecule cases use different input conformers. Therefore these plots show observed differences, not isolated causal effects of removing constraints.
55
+
56
+ Historical originals with unavailable separate native/raw/AF3 artifacts remain represented. `artifacts-audited.csv` and `input_coverage.csv` distinguish available files from unresolved or intentionally retired evidence; a missing file is not a new failed design.
benchmarks/provenance.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "source_release": "supplement-20261009",
3
+ "paper_commit": "8614542c149875b13ed3fed5d8d6ec739f2daff4",
4
+ "source_path": "paper/figures/figure 2/benchmark_packet/supplement",
5
+ "metric": "mean_over_targets(3600 * passing_structure_clusters / (generation_gpu_s + af3_gpu_s))",
6
+ "summary_rows_recomputed": 52,
7
+ "target_criterion_rows": 2488,
8
+ "sha256": {
9
+ "summary.csv": "5c6d525d988d0df7fecb74396f86401c1ddb4c31fa83f00b357ba77d855d8a10",
10
+ "source_data.csv": "5f71454272467ea86a5ef0f4ad8a7c781ca31eff8d57497e83c87aa8820cc34b"
11
+ },
12
+ "scope": "Frozen project benchmark runs; not a new evaluation of every model Hub revision."
13
+ }
benchmarks/source_data.csv ADDED
The diff for this file is too large to render. See raw diff
 
benchmarks/summary.csv ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ panel,model,protocol,dataset,label,criterion,pass_percent,clustered_percent,successes_per_10_gpu_min,unique_per_10_gpu_min,targets,attempts,timing_note
2
+ ame,alphaprotein_novo,one_shot,mcsa_fixed,AlphaProtein-novo (Codesign),ame,7.986486486486487,5.675675675675675,6.128165410759252,4.336287958764792,37,7400,per_attempt
3
+ ame,alphaprotein_novo_ligandmpnn,one_shot,mcsa_fixed,AlphaProtein-novo (LigandMPNN),ame,7.418918918918919,5.418918918918919,5.699414945101433,4.143206492049179,37,7400,inherited_parent;inherited_parent_generation_CPU_MPNN
4
+ ame,odesign,one_shot,mcsa_fixed,ODesign,ame,0.9054054054054054,0.2702702702702703,0.6290490961497859,0.1834221580606205,37,7400,per_attempt;original_generation_timing_proxy_v1
5
+ ame,pipelines,one_shot,mcsa_fixed,RFdiffusion3,ame,2.864864864864865,2.27027027027027,3.6288471843745116,2.873393749308303,37,7400,per_attempt
6
+ ame,proteina_complexa,one_shot,mcsa_fixed,Proteina-Complexa,ame,3.283783783783784,2.324324324324324,6.040088839302828,4.217084980960173,37,7400,per_attempt
7
+ ame,rfd4_proteina_high,one_shot,mcsa_fixed,RFD4-Proteina (high thinking),ame,13.445945945945946,8.905405405405405,23.248249750649716,15.197139241828658,37,7400,per_attempt
8
+ ame,rfd4_proteina_high_search,native_pipeline,mcsa_fixed,RFD4-Proteina (search),ame,17.81081081081081,12.432432432432432,2.1139479255711926,1.4702629167109056,37,7400,per_attempt
9
+ ame,rfd4_proteina_low,one_shot,mcsa_fixed,RFD4-Proteina (low thinking),ame,10.62162162162162,7.108108108108108,24.61903561723674,16.21640377305806,37,7400,per_attempt
10
+ ame,rfd4_proteina_medium,one_shot,mcsa_fixed,RFD4-Proteina (medium thinking),ame,12.37837837837838,8.216216216216216,26.60857770347569,17.39937846539931,37,7400,per_attempt
11
+ ame,rfdiffusion2,one_shot,mcsa_fixed,RFdiffusion2,ame,0.8378378378378378,0.7837837837837838,0.054484078041217035,0.05061505532429724,37,7400,per_attempt
12
+ dna,boltzgen,one_shot,dna_fixed,BoltzGen (fixed),dna,0.8333333333333334,0.8333333333333334,1.5767861157347076,1.5767861157347076,3,600,per_attempt
13
+ dna,odesign,one_shot,dna_fixed,ODesign (fixed),dna,0.0,0.0,0.0,0.0,3,600,per_attempt
14
+ dna,pipelines,one_shot,dna_fixed,RFdiffusion3 (fixed),dna,0.1666666666666666,0.1666666666666666,0.12549701345123473,0.12549701345123473,3,600,per_attempt
15
+ dna,protein_hunter,native_pipeline,dna_unfixed,Protein-Hunter,dna,0.3333333333333333,0.3333333333333333,0.0339968718491103,0.0339968718491103,3,600,per_attempt
16
+ dna,rfd4_proteina_high,one_shot,dna_unfixed,RFD4-Proteina (high thinking),dna,2.333333333333333,2.1666666666666665,7.533663081020337,6.995141754963048,3,600,per_attempt
17
+ dna,rfd4_proteina_high_search,native_pipeline,dna_unfixed,RFD4-Proteina (search),dna,3.8333333333333335,3.5,0.6977291779821485,0.637527788137053,3,600,per_attempt
18
+ dna,rfd4_proteina_low,one_shot,dna_unfixed,RFD4-Proteina (low thinking),dna,1.3333333333333333,1.3333333333333333,4.501005002182411,4.501005002182411,3,600,per_attempt
19
+ dna,rfd4_proteina_medium,one_shot,dna_unfixed,RFD4-Proteina (medium thinking),dna,2.0,1.6666666666666667,6.57941016603549,5.484803149401414,3,600,per_attempt
20
+ ppi,bindcraft,one_shot,ppi_all_atom,BindCraft (single backbone),ppi,9.9375,5.854166666666667,0.6858558193759934,0.38947020233767843,24,4800,per_attempt
21
+ ppi,bindcraft2,one_shot,ppi_all_atom,BindCraft2 (single backbone),ppi,7.229166666666667,4.8125,0.5281467210933602,0.3331281596040387,24,4800,per_attempt
22
+ ppi,boltzgen,one_shot,ppi_all_atom,BoltzGen,ppi,4.8125,1.2916666666666667,5.406141349312153,1.3833348139931438,24,4800,per_attempt
23
+ ppi,esmfold2,one_shot,ppi_backbone,ESMFold2,ppi,8.916666666666666,2.458333333333333,0.7509894473902138,0.19387923161633483,24,4800,per_attempt
24
+ ppi,genie3,one_shot,ppi_all_atom,Genie3,ppi,0.6875,0.5208333333333334,0.5547110110121317,0.42859016992113674,24,4800,per_attempt
25
+ ppi,odesign,one_shot,ppi_all_atom,ODesign,ppi,1.4583333333333333,1.2916666666666667,0.7635954268204376,0.668254791590555,24,4800,per_attempt;original_generation_timing_proxy_v1
26
+ ppi,pipelines,one_shot,ppi_all_atom,RFdiffusion3,ppi,3.375,1.7916666666666667,2.203438218078908,1.1140170381975611,24,4800,per_attempt
27
+ ppi,protein_hunter,native_pipeline,ppi_all_atom,Protein-Hunter,ppi,6.916666666666667,4.145833333333333,0.5573716351941015,0.33134699754642055,24,4800,per_attempt
28
+ ppi,proteina_complexa,one_shot,ppi_all_atom,Proteina-Complexa,ppi,4.5,1.8541666666666667,5.446523655582822,2.207096244584333,24,4800,per_attempt
29
+ ppi,rfd4-search-multimer,native_pipeline,ppi_backbone,RFD4-Proteina (search),ppi,24.45833333333333,11.916666666666666,1.7550251894828512,0.8465949585039426,24,4800,per_attempt
30
+ ppi,rfd4_proteina_high,one_shot,ppi_backbone,RFD4-Proteina (high thinking),ppi,19.14583333333333,9.770833333333334,24.50772631322702,11.933874547020181,24,4800,batch_amortized;batched_generation_remeasurement_v1
31
+ ppi,rfd4_proteina_low,one_shot,ppi_backbone,RFD4-Proteina (low thinking),ppi,14.104166666666666,7.041666666666667,23.666769202318935,11.560507426898935,24,4800,batch_amortized;batched_generation_remeasurement_v1
32
+ ppi,rfd4_proteina_medium,one_shot,ppi_backbone,RFD4-Proteina (medium thinking),ppi,16.395833333333332,8.354166666666666,25.812263933585978,12.73345240158931,24,4800,batch_amortized;batched_generation_remeasurement_v1
33
+ ppi,rfdiffusion,one_shot,ppi_all_atom,RFdiffusion,ppi,2.6041666666666665,0.75,0.5691347855847814,0.15660421612078013,24,4800,per_attempt
34
+ small_molecule,boltzgen,one_shot,studio_figure_2c_unfixed,BoltzGen,sm_binder,4.045454545454546,1.318181818181818,8.145222361450198,2.579619122276541,77,15400,per_attempt
35
+ small_molecule,disco,one_shot,studio_figure_2c_unfixed,DISCO,sm_binder,2.551948051948052,1.5714285714285714,0.20401419810891713,0.12533260451848793,77,15400,per_attempt
36
+ small_molecule,odesign,one_shot,studio_figure_2c_fixed,ODesign (fixed),sm_binder,0.0974025974025974,0.0519480519480519,0.06812084076412611,0.0366360091773046,77,15400,per_attempt
37
+ small_molecule,pipelines,one_shot,studio_figure_2c_unfixed,RFdiffusion3,sm_binder,3.512987012987013,2.3376623376623376,3.2330633144392786,2.1472317944116814,77,15400,per_attempt
38
+ small_molecule,protein_hunter,native_pipeline,studio_figure_2c_unfixed,Protein-Hunter,sm_binder,0.7857142857142857,0.6688311688311688,0.08222397901207588,0.06994325849054965,77,15400,per_attempt
39
+ small_molecule,proteina_complexa,one_shot,studio_figure_2c_fixed,Proteina-Complexa (fixed),sm_binder,0.1493506493506493,0.1428571428571428,0.3550431740456386,0.3394022404125218,77,15400,batch_amortized;batched_generation_remeasurement_v1
40
+ small_molecule,rfd4_proteina_high,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (high thinking),sm_binder,27.818181818181817,8.74025974025974,59.80580730283336,18.708070088438323,77,15400,per_attempt
41
+ small_molecule,rfd4_proteina_high_search,native_pipeline,studio_figure_2c_unfixed,RFD4-Proteina (search),sm_binder,40.29220779220779,12.863636363636363,6.4266313688585415,2.042660497770513,77,15400,per_attempt
42
+ small_molecule,rfd4_proteina_low,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (low thinking),sm_binder,20.92207792207792,6.37012987012987,58.905492101587086,17.847324331293226,77,15400,per_attempt
43
+ small_molecule,rfd4_proteina_medium,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (medium thinking),sm_binder,25.18831168831169,7.577922077922078,66.53565733313984,19.911485476039143,77,15400,per_attempt
44
+ small_molecule_original,boltzgen,one_shot,studio_figure_2c_unfixed,BoltzGen,disco,27.675324675324678,6.097402597402597,53.78792519211063,11.761316579129796,77,15400,per_attempt
45
+ small_molecule_original,disco,one_shot,studio_figure_2c_unfixed,DISCO,disco,31.32467532467533,10.474025974025974,2.4539028999553714,0.8175381293508085,77,15400,per_attempt
46
+ small_molecule_original,odesign,one_shot,studio_figure_2c_fixed,ODesign (fixed),disco,1.5909090909090908,1.025974025974026,1.1154226419662379,0.7190330690062327,77,15400,per_attempt
47
+ small_molecule_original,pipelines,one_shot,studio_figure_2c_unfixed,RFdiffusion3,disco,26.13636363636364,12.844155844155845,23.779749576062088,11.679276621280051,77,15400,per_attempt
48
+ small_molecule_original,protein_hunter,native_pipeline,studio_figure_2c_unfixed,Protein-Hunter,disco,3.961038961038961,2.415584415584416,0.41375915688116066,0.2519214097355728,77,15400,per_attempt
49
+ small_molecule_original,proteina_complexa,one_shot,studio_figure_2c_fixed,Proteina-Complexa (fixed),disco,14.818181818181818,12.175324675324676,32.58468485951315,26.699142756356586,77,15400,batch_amortized;batched_generation_remeasurement_v1
50
+ small_molecule_original,rfd4_proteina_high,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (high thinking),disco,63.25974025974026,13.915584415584416,136.2596016660901,29.843032429691466,77,15400,per_attempt
51
+ small_molecule_original,rfd4_proteina_high_search,native_pipeline,studio_figure_2c_unfixed,RFD4-Proteina (search),disco,73.35064935064935,18.11688311688312,11.821053066769755,2.892460870734589,77,15400,per_attempt
52
+ small_molecule_original,rfd4_proteina_low,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (low thinking),disco,57.18181818181818,11.915584415584416,160.78300033866194,33.311256689662514,77,15400,per_attempt
53
+ small_molecule_original,rfd4_proteina_medium,one_shot,studio_figure_2c_unfixed,RFD4-Proteina (medium thinking),disco,61.0974025974026,12.474025974025974,161.40828898514403,32.75934584692291,77,15400,per_attempt
config.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "rfd4_proteina",
3
+ "format_version": 1,
4
+ "weights_sha256": "5b549dd61adb2f6c6b5cb95bda13f40f5e31556b4e6c755edc0ca6f011c68fb0",
5
+ "config": {
6
+ "nn": {
7
+ "name": "local_latents_transformer",
8
+ "output_parameterization": {
9
+ "bb_ca": "v",
10
+ "local_latents": "v"
11
+ },
12
+ "sequence_readout": {
13
+ "enabled": true,
14
+ "detach": true,
15
+ "num_classes": 147
16
+ },
17
+ "nlayers": 14,
18
+ "token_dim": 768,
19
+ "nheads": 12,
20
+ "parallel_mha_transition": false,
21
+ "token_type_emb_dim": 32,
22
+ "cath_emb_dim": 32,
23
+ "strict_feats": false,
24
+ "noise_align_chirality": true,
25
+ "noise_align_internal_coords": false,
26
+ "noise_align_chirality_translate": false,
27
+ "noise_align_internal_coords_translate": false,
28
+ "use_atom37_factory": true,
29
+ "feats_seq": [
30
+ "xt_bb_ca",
31
+ "xt_local_latents",
32
+ "token_type",
33
+ "sequence_probabilities",
34
+ "cond_a37coors_nm",
35
+ "cond_a37coors_internal_nm",
36
+ "rasa_binned",
37
+ "secondary_structure",
38
+ "hbond_modernized",
39
+ "pistack",
40
+ "pication",
41
+ "saltbridge",
42
+ "hydrophobic",
43
+ "metal_coord",
44
+ "na_ss_state",
45
+ "hotspot",
46
+ "atom_charge",
47
+ "atomic_element",
48
+ "coord_group_seq",
49
+ "chirality",
50
+ "chain_centroid_xyz"
51
+ ],
52
+ "feats_cond_seq": [
53
+ "time_emb_bb_ca",
54
+ "time_emb_local_latents",
55
+ "cath_code",
56
+ "recycle_remaining"
57
+ ],
58
+ "dim_cond": 256,
59
+ "idx_emb_dim": 256,
60
+ "t_emb_dim": 256,
61
+ "feats_pair_repr": [
62
+ "xt_bb_ca_pair_dists",
63
+ "bond_mask",
64
+ "bond_order",
65
+ "is_aromatic",
66
+ "index_group_pair",
67
+ "molecule_idx_pair",
68
+ "molecule_entity_pair",
69
+ "token_bond_path_length",
70
+ "canonical_polymer_ordering",
71
+ "na_ss_pair"
72
+ ],
73
+ "feats_pair_cond": [
74
+ "time_emb_bb_ca",
75
+ "time_emb_local_latents"
76
+ ],
77
+ "xt_pair_dist_dim": 30,
78
+ "xt_pair_dist_min": 0.1,
79
+ "xt_pair_dist_max": 3,
80
+ "seq_sep_dim": 64,
81
+ "pair_repr_dim": 384,
82
+ "update_pair_repr": true,
83
+ "update_pair_repr_every_n": 2,
84
+ "use_tri_mult": true,
85
+ "tri_mult_c": 256,
86
+ "tri_mult_bias": false,
87
+ "tri_mult_backend": "cuequivariance",
88
+ "recycle": {
89
+ "enabled": true,
90
+ "grad": "last",
91
+ "training_max_recycles": 9,
92
+ "condition_probability": 0.5
93
+ },
94
+ "checkpointing": "none",
95
+ "pair_write": "hadamard",
96
+ "use_qkln": true,
97
+ "attention_type": "torch",
98
+ "use_torch_compile": false,
99
+ "norm_type": "rmsnorm_torch_fused"
100
+ },
101
+ "product_flowmatcher": {
102
+ "bb_ca": {
103
+ "zero_com_noise": true,
104
+ "guidance_enabled": true,
105
+ "dim": 3,
106
+ "stochastic_centering_mode": "x_0",
107
+ "stochastic_centering_scale": 0.2,
108
+ "stochastic_centering_scope": "per_group",
109
+ "stochastic_centering_group_keys": [
110
+ "molecule_iid"
111
+ ]
112
+ },
113
+ "local_latents": {
114
+ "zero_com_noise": false,
115
+ "guidance_enabled": true,
116
+ "dim": 32,
117
+ "stochastic_centering_mode": "none",
118
+ "stochastic_centering_scale": null
119
+ }
120
+ },
121
+ "loss": {
122
+ "signed_volume_weight": 2.5,
123
+ "sequence_prediction": {
124
+ "enabled": true,
125
+ "t_min": 0.0,
126
+ "t_max": 1.0
127
+ },
128
+ "t_distribution": {
129
+ "bb_ca": {
130
+ "name": "mix_unif_beta",
131
+ "p1": 1.9,
132
+ "p2": 1.0,
133
+ "p3": 0.02
134
+ },
135
+ "local_latents": {
136
+ "name": "mix_unif_beta",
137
+ "p1": 1.0,
138
+ "p2": 1.5,
139
+ "p3": 0.2
140
+ },
141
+ "shared_groups": null
142
+ }
143
+ },
144
+ "training": {
145
+ "exploration_k": 4,
146
+ "exploration_losses": [
147
+ "bb_ca",
148
+ "local_latents"
149
+ ],
150
+ "n_recycle_max": 9,
151
+ "n_recycle_sampler": "poisson",
152
+ "n_recycle_lambda": 3.0,
153
+ "n_recycle_sync": true,
154
+ "n_recycle_val_loss": 3,
155
+ "motif_conditioning": false
156
+ },
157
+ "opt": {
158
+ "lr": 5e-05,
159
+ "muon_warmup_steps": 0,
160
+ "adam_warmup_steps": 10000,
161
+ "gradient_clip_val": 1.0,
162
+ "max_epochs": 10000000,
163
+ "accumulate_grad_batches": 2,
164
+ "dist_strategy": "ddp",
165
+ "steps_per_epoch": 1000,
166
+ "val_every_n_epochs": 5,
167
+ "val_start_epoch": 0,
168
+ "compute_val_loss": false,
169
+ "foldseek_timeout_minutes": 25,
170
+ "muon_exclude_param_patterns": [
171
+ "_factory\\.",
172
+ "^nn\\.(local_latents_linear|ca_linear|insertion_q_proj|insertion_k_proj)\\.",
173
+ "^nn\\.sequence_readout\\."
174
+ ],
175
+ "muon_adjust_lr_fn": "match_rms_adamw",
176
+ "muon_grouped": true,
177
+ "num_sanity_val_steps": 2,
178
+ "coordinate_descent_tuning": false
179
+ },
180
+ "autoencoder_config": {
181
+ "nn_ae": {
182
+ "name": "transformer_ae_aa_rna_dna_glycan_ncaa",
183
+ "latent_z_dim": 32,
184
+ "allow_oxt": true,
185
+ "use_torch_compile": false,
186
+ "torch_compile_fullgraph": false,
187
+ "n_residue_types": 147,
188
+ "encoder": {
189
+ "nlayers": 12,
190
+ "token_dim": 768,
191
+ "nheads": 12,
192
+ "parallel_mha_transition": false,
193
+ "token_type_emb_dim": 32,
194
+ "strict_feats": false,
195
+ "feats_seq": [
196
+ "token_type",
197
+ "x1_a37coors_nm",
198
+ "x1_a37coors_nm_rel",
199
+ "chain_idx_seq",
200
+ "incident_bond_type_counts",
201
+ "atom37_element_charge_embedding"
202
+ ],
203
+ "feats_cond_seq": null,
204
+ "dim_cond": 128,
205
+ "idx_emb_dim": 128,
206
+ "feats_pair_repr": [
207
+ "token_bond_path_length",
208
+ "canonical_polymer_ordering",
209
+ "molecule_idx_pair",
210
+ "x1_bb_pair_dists_nm"
211
+ ],
212
+ "seq_sep_dim": 64,
213
+ "pair_repr_dim": 256,
214
+ "update_pair_repr": false,
215
+ "update_pair_repr_every_n": 3,
216
+ "use_tri_mult": false,
217
+ "use_attn_pair_bias": false,
218
+ "use_qkln": true,
219
+ "normalize_latent": false,
220
+ "latent_z_dim": 32,
221
+ "n_residue_types": 147,
222
+ "charge_emb_dim": 16
223
+ },
224
+ "decoder": {
225
+ "nlayers": 12,
226
+ "token_dim": 768,
227
+ "abs_coors": false,
228
+ "nheads": 12,
229
+ "token_type_emb_dim": 32,
230
+ "parallel_mha_transition": false,
231
+ "strict_feats": false,
232
+ "feats_seq": [
233
+ "ca_coors_nm",
234
+ "z_latent_seq",
235
+ "token_type",
236
+ "cond_residue_type"
237
+ ],
238
+ "feats_cond_seq": null,
239
+ "dim_cond": 128,
240
+ "idx_emb_dim": 128,
241
+ "feats_pair_repr": [
242
+ "token_bond_path_length",
243
+ "canonical_polymer_ordering",
244
+ "molecule_idx_pair",
245
+ "ca_coors_nm_pair_dists"
246
+ ],
247
+ "seq_sep_dim": 64,
248
+ "pair_repr_dim": 256,
249
+ "update_pair_repr": false,
250
+ "update_pair_repr_every_n": 3,
251
+ "use_tri_mult": false,
252
+ "use_attn_pair_bias": false,
253
+ "use_qkln": true,
254
+ "latent_z_dim": 32,
255
+ "n_residue_types": 147
256
+ }
257
+ }
258
+ },
259
+ "path_length_search_cutoffs": [
260
+ 64
261
+ ]
262
+ }
263
+ }
examples/antibody_design/design.cif ADDED
The diff for this file is too large to render. See raw diff
 
examples/antibody_design/input.cif ADDED
The diff for this file is too large to render. See raw diff
 
examples/enzyme_design/design.cif ADDED
The diff for this file is too large to render. See raw diff
 
examples/enzyme_design/input.cif ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ data_M0630_1j79.cif
2
+ #
3
+ _entry.id M0630_1j79.cif
4
+ _entry.author rib7
5
+ _entry.date 2026-08-28
6
+ _entry.time 20:38:54
7
+ #
8
+ loop_
9
+ _entity.id
10
+ _entity.type
11
+ _entity.pdbx_description
12
+ 0 polymer .
13
+ 1 non-polymer .
14
+ 2 non-polymer .
15
+ #
16
+ _entity_poly.entity_id 0
17
+ _entity_poly.type polypeptide(L)
18
+ _entity_poly.nstd_linkage no
19
+ _entity_poly.nstd_monomer yes
20
+ _entity_poly.pdbx_seq_one_letter_code XD
21
+ _entity_poly.pdbx_seq_one_letter_code_can XD
22
+ _entity_poly.pdbx_strand_id A
23
+ _entity_poly.pdbx_target_identifier ?
24
+ #
25
+ loop_
26
+ _entity_poly_seq.entity_id
27
+ _entity_poly_seq.hetero
28
+ _entity_poly_seq.mon_id
29
+ _entity_poly_seq.num
30
+ 0 n 180-180 0
31
+ 0 n ASP 1
32
+ #
33
+ loop_
34
+ _chem_comp.id
35
+ _chem_comp.name
36
+ _chem_comp.type
37
+ _chem_comp.formula
38
+ _chem_comp.mon_nstd_parent_comp_id
39
+ _chem_comp.pdbx_synonyms
40
+ _chem_comp.formula_weight
41
+ 180-180 180-180 NON-POLYMER ? ? ? ?
42
+ ASP ASP 'L-PEPTIDE LINKING' ? ? ? ?
43
+ ORO ORO NON-POLYMER ? ? ? ?
44
+ ZN ZN NON-POLYMER ? ? ? ?
45
+ #
46
+ loop_
47
+ _chem_comp_bond.comp_id
48
+ _chem_comp_bond.atom_id_1
49
+ _chem_comp_bond.atom_id_2
50
+ _chem_comp_bond.value_order
51
+ _chem_comp_bond.pdbx_aromatic_flag
52
+ ASP N CA SING N
53
+ ASP N H SING N
54
+ ASP N H2 SING N
55
+ ASP CA C SING N
56
+ ASP CA CB SING N
57
+ ASP CA HA SING N
58
+ ASP C O DOUB N
59
+ ASP C OXT SING N
60
+ ASP CB CG SING N
61
+ ASP CB HB2 SING N
62
+ ASP CB HB3 SING N
63
+ ASP CG OD1 DOUB N
64
+ ASP CG OD2 SING N
65
+ ASP OD2 HD2 SING N
66
+ ASP OXT HXT SING N
67
+ ORO N1 C2 SING Y
68
+ ORO N1 C6 SING Y
69
+ ORO N1 HN1 SING N
70
+ ORO C2 O2 DOUB N
71
+ ORO C2 N3 SING Y
72
+ ORO N3 C4 SING Y
73
+ ORO N3 HN3 SING N
74
+ ORO C4 O4 DOUB N
75
+ ORO C4 C5 SING Y
76
+ ORO C5 C6 DOUB Y
77
+ ORO C5 H5 SING N
78
+ ORO C6 C7 SING N
79
+ ORO C7 O71 DOUB N
80
+ ORO C7 O72 SING N
81
+ ORO O72 HO7 SING N
82
+ #
83
+ loop_
84
+ _atom_site.group_PDB
85
+ _atom_site.type_symbol
86
+ _atom_site.label_atom_id
87
+ _atom_site.label_alt_id
88
+ _atom_site.label_comp_id
89
+ _atom_site.label_asym_id
90
+ _atom_site.label_entity_id
91
+ _atom_site.label_seq_id
92
+ _atom_site.pdbx_PDB_ins_code
93
+ _atom_site.auth_seq_id
94
+ _atom_site.auth_comp_id
95
+ _atom_site.auth_asym_id
96
+ _atom_site.auth_atom_id
97
+ _atom_site.B_iso_or_equiv
98
+ _atom_site.occupancy
99
+ _atom_site.pdbx_formal_charge
100
+ _atom_site.annotation_src-component_1_atom
101
+ _atom_site.annotation_expsegmax_1_residue
102
+ _atom_site.mask_expsegmin_1_residue
103
+ _atom_site.condition_hbond_1_atom
104
+ _atom_site.condition_index-group_1_token
105
+ _atom_site.annotation_expsegmin_1_residue
106
+ _atom_site.mask_index_1_residue
107
+ _atom_site.mask_sequence_1_residue
108
+ _atom_site.mask_expsegmax_1_residue
109
+ _atom_site.mask_coordinate_1_atom
110
+ _atom_site.atom_array_index
111
+ _atom_site.Cartn_x
112
+ _atom_site.Cartn_y
113
+ _atom_site.Cartn_z
114
+ _atom_site.pdbx_PDB_model_num
115
+ _atom_site.id
116
+ HETATM X UNK . 180-180 A 0 0 . 0 180-180 A UNK nan 0.0 ? . 180 True 0 0 180 True False True False 0 nan nan nan 1 1
117
+ ATOM C CG . ASP A 0 1 . 1 ASP A CG 17.39 1.0 ? A250 -1 False 0 4 -1 False True False True 1 0.477 -0.542 2.937 1 2
118
+ ATOM O OD1 . ASP A 0 1 . 1 ASP A OD1 18.72 1.0 ? A250 -1 False 0 4 -1 False True False True 2 0.547 0.656 2.665 1 3
119
+ HETATM N N1 . ORO C 2 410 . 410 ORO C N1 18.66 1.0 ? ORO -1 False 0 1 -1 True True False True 3 -2.127 0.345 -1.747 1 4
120
+ HETATM C C2 . ORO C 2 410 . 410 ORO C C2 23.79 1.0 ? ORO -1 False 0 1 -1 True True False True 4 -1.547 -0.999 -1.559 1 5
121
+ HETATM O O2 . ORO C 2 410 . 410 ORO C O2 19.98 1.0 ? ORO -1 False 0 1 -1 True True False False 5 0.0 0.0 0.0 1 6
122
+ HETATM N N3 . ORO C 2 410 . 410 ORO C N3 26.76 1.0 ? ORO -1 False 0 1 -1 True True False True 6 -0.172 -1.019 -1.547 1 7
123
+ HETATM C C4 . ORO C 2 410 . 410 ORO C C4 100.0 1.0 ? ORO -1 False 0 1 -1 True True False True 7 0.602 -0.003 -2.126 1 8
124
+ HETATM O O4 . ORO C 2 410 . 410 ORO C O4 39.79 1.0 ? ORO -1 False 0 1 -1 True True False False 8 0.0 0.0 0.0 1 9
125
+ HETATM C C5 . ORO C 2 410 . 410 ORO C C5 30.44 1.0 ? ORO -1 False 0 1 -1 True True False False 9 0.0 0.0 0.0 1 10
126
+ HETATM C C6 . ORO C 2 410 . 410 ORO C C6 29.02 1.0 ? ORO -1 False 0 1 -1 True True False False 10 0.0 0.0 0.0 1 11
127
+ HETATM C C7 . ORO C 2 410 . 410 ORO C C7 21.02 1.0 ? ORO -1 False 0 1 -1 True True False False 11 0.0 0.0 0.0 1 12
128
+ HETATM O O71 . ORO C 2 410 . 410 ORO C O71 17.46 1.0 ? ORO -1 False 0 1 -1 True True False False 12 0.0 0.0 0.0 1 13
129
+ HETATM O O72 . ORO C 2 410 . 410 ORO C O72 18.24 1.0 ? ORO -1 False 0 1 -1 True True False False 13 0.0 0.0 0.0 1 14
130
+ HETATM ZN ZN . ZN B 1 400 . 400 ZN B ZN 22.02 1.0 +2 ZN -1 False 0 2 -1 True True False True 14 2.223 1.561 1.378 1 15
131
+ #
132
+ _rfd_design.sampled_contig 180-180
133
+ #
134
+ loop_
135
+ _chem_comp_atom.comp_id
136
+ _chem_comp_atom.atom_id
137
+ _chem_comp_atom.type_symbol
138
+ _chem_comp_atom.charge
139
+ _chem_comp_atom.pdbx_aromatic_flag
140
+ _chem_comp_atom.pdbx_leaving_atom_flag
141
+ _chem_comp_atom.pdbx_model_Cartn_x_ideal
142
+ _chem_comp_atom.pdbx_model_Cartn_y_ideal
143
+ _chem_comp_atom.pdbx_model_Cartn_z_ideal
144
+ ASP N N 0 N N -0.31700000166893005 1.687999963760376 0.06599999964237213
145
+ ASP CA C 0 N N -0.4699999988079071 0.28600001335144043 -0.3440000116825104
146
+ ASP C C 0 N N -1.8680000305175781 -0.18000000715255737 -0.028999999165534973
147
+ ASP O O 0 N N -2.5339999198913574 0.41499999165534973 0.7860000133514404
148
+ ASP CB C 0 N N 0.5389999747276306 -0.5799999833106995 0.4129999876022339
149
+ ASP CG C 0 N N 1.937999963760376 -0.19499999284744263 0.004000000189989805
150
+ ASP OD1 O 0 N N 2.1089999675750732 0.6809999942779541 -0.8100000023841858
151
+ ASP OD2 O 0 N N 2.992000102996826 -0.8259999752044678 0.5429999828338623
152
+ ASP OXT O 0 N Y -2.374000072479248 -1.25600004196167 -0.6520000100135803
153
+ ASP H H 0 N N -0.9279999732971191 2.2890000343322754 -0.46700000762939453
154
+ ASP H2 H 0 N Y -0.4779999852180481 1.7949999570846558 1.055999994277954
155
+ ASP HA H 0 N N -0.2919999957084656 0.19900000095367432 -1.4160000085830688
156
+ ASP HB2 H 0 N N 0.4189999997615814 -0.42500001192092896 1.4850000143051147
157
+ ASP HB3 H 0 N N 0.367000013589859 -1.6299999952316284 0.17599999904632568
158
+ ASP HD2 H 0 N N 3.86899995803833 -0.5450000166893005 0.25
159
+ ASP HXT H 0 N Y -3.2750000953674316 -1.5169999599456787 -0.41600000858306885
160
+ ORO N1 N 0 Y N 1.1260000467300415 0.004999999888241291 0.041999999433755875
161
+ ORO C2 C 0 Y N 1.1929999589920044 -0.0010000000474974513 -1.2990000247955322
162
+ ORO O2 O 0 N N 2.2839999198913574 -0.0010000000474974513 -1.8359999656677246
163
+ ORO N3 N 0 Y N 0.08799999952316284 -0.0 -2.062999963760376
164
+ ORO C4 C 0 Y N -1.1349999904632568 -0.0 -1.4950000047683716
165
+ ORO O4 O 0 N N -2.138000011444092 0.0 -2.184000015258789
166
+ ORO C5 C 0 Y N -1.2359999418258667 -0.0 -0.029999999329447746
167
+ ORO C6 C 0 Y N -0.08699999749660492 -0.0 0.6949999928474426
168
+ ORO C7 C 0 N N -0.13899999856948853 -0.0 2.177000045776367
169
+ ORO O71 O 0 N N -1.2109999656677246 -0.0 2.746000051498413
170
+ ORO O72 O 0 N N 1.0019999742507935 -0.0 2.8910000324249268
171
+ ORO HN1 H 0 N N 1.9450000524520874 0.004000000189989805 0.5609999895095825
172
+ ORO HN3 H 0 N N 0.17000000178813934 -0.0 -3.0299999713897705
173
+ ORO H5 H 0 N N -2.1989998817443848 -0.0 0.4569999873638153
174
+ ORO HO7 H 0 N N 0.9679999947547913 -0.0 3.8570001125335693
175
+ ZN ZN ZN 2 N N nan nan nan
176
+ 180-180 UNK X 0 N N nan nan nan
177
+ #
examples/monomer.cif ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ data_monomer-short.cif
2
+ #
3
+ _entry.id monomer-short.cif
4
+ _entry.author rib7
5
+ _entry.date 2026-08-28
6
+ _entry.time 19:44:50
7
+ #
8
+ _entity.id 0
9
+ _entity.type polymer
10
+ _entity.pdbx_description .
11
+ #
12
+ _entity_poly.entity_id 0
13
+ _entity_poly.type polypeptide(L)
14
+ _entity_poly.nstd_linkage no
15
+ _entity_poly.nstd_monomer yes
16
+ _entity_poly.pdbx_seq_one_letter_code X
17
+ _entity_poly.pdbx_seq_one_letter_code_can X
18
+ _entity_poly.pdbx_strand_id A
19
+ _entity_poly.pdbx_target_identifier ?
20
+ #
21
+ _entity_poly_seq.entity_id 0
22
+ _entity_poly_seq.hetero n
23
+ _entity_poly_seq.mon_id 100-100
24
+ _entity_poly_seq.num 1
25
+ #
26
+ _chem_comp.id 100-100
27
+ _chem_comp.name 100-100
28
+ _chem_comp.type NON-POLYMER
29
+ _chem_comp.formula ?
30
+ _chem_comp.mon_nstd_parent_comp_id ?
31
+ _chem_comp.pdbx_synonyms ?
32
+ _chem_comp.formula_weight ?
33
+ #
34
+ _atom_site.group_PDB HETATM
35
+ _atom_site.type_symbol X
36
+ _atom_site.label_atom_id UNK
37
+ _atom_site.label_alt_id .
38
+ _atom_site.label_comp_id 100-100
39
+ _atom_site.label_asym_id A
40
+ _atom_site.label_entity_id 0
41
+ _atom_site.label_seq_id 1
42
+ _atom_site.pdbx_PDB_ins_code .
43
+ _atom_site.auth_seq_id 1
44
+ _atom_site.auth_comp_id 100-100
45
+ _atom_site.auth_asym_id A
46
+ _atom_site.auth_atom_id UNK
47
+ _atom_site.B_iso_or_equiv nan
48
+ _atom_site.occupancy 0.0
49
+ _atom_site.pdbx_formal_charge ?
50
+ _atom_site.annotation_expsegmax_1_residue 100
51
+ _atom_site.mask_expsegmin_1_residue True
52
+ _atom_site.mask_expsegmax_1_residue True
53
+ _atom_site.annotation_expsegmin_1_residue 100
54
+ _atom_site.condition_hbond_1_atom 0
55
+ _atom_site.condition_index-group_1_token 0
56
+ _atom_site.mask_index_1_residue True
57
+ _atom_site.mask_coordinate_1_atom False
58
+ _atom_site.mask_sequence_1_residue False
59
+ _atom_site.annotation_src-component_1_atom .
60
+ _atom_site.atom_array_index 0
61
+ _atom_site.Cartn_x nan
62
+ _atom_site.Cartn_y nan
63
+ _atom_site.Cartn_z nan
64
+ _atom_site.pdbx_PDB_model_num 1
65
+ _atom_site.id 1
66
+ #
67
+ _rfd_design.sampled_contig 100-100
68
+ #
69
+ _chem_comp_atom.comp_id 100-100
70
+ _chem_comp_atom.atom_id UNK
71
+ _chem_comp_atom.type_symbol X
72
+ _chem_comp_atom.charge 0
73
+ _chem_comp_atom.pdbx_aromatic_flag N
74
+ _chem_comp_atom.pdbx_leaving_atom_flag N
75
+ _chem_comp_atom.pdbx_model_Cartn_x_ideal nan
76
+ _chem_comp_atom.pdbx_model_Cartn_y_ideal nan
77
+ _chem_comp_atom.pdbx_model_Cartn_z_ideal nan
78
+ #
examples/protein_binder/design.cif ADDED
The diff for this file is too large to render. See raw diff
 
examples/protein_binder/input.cif ADDED
The diff for this file is too large to render. See raw diff
 
flow_checkpoints/rfd4_proteina_flow-ema.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1c8b7068e87c41a334ec5b77e7d7cdda30bd31ed4504909d93bd72f7742afe4
3
+ size 1751488703
flow_checkpoints/rfd4_proteina_flow-ema.ckpt.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ c1c8b7068e87c41a334ec5b77e7d7cdda30bd31ed4504909d93bd72f7742afe4 rfd4_proteina_flow-ema.ckpt
flow_checkpoints/rfd4_proteina_flow_pretrain-ema.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:561bbbe3e97f37ded11ddc934ff2184adcd98038396f7aa1f16e6f9d57f0f8b2
3
+ size 1743470486
flow_checkpoints/rfd4_proteina_flow_pretrain-ema.ckpt.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 561bbbe3e97f37ded11ddc934ff2184adcd98038396f7aa1f16e6f9d57f0f8b2 rfd4_proteina_flow_pretrain-ema.ckpt
inference.yaml ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Default configs for inference through generate.py
2
+ # The defaults are geared more towards ease-of-use by designers, so running with
3
+ # the existing analysis scripts will necessitate overrides
4
+
5
+ defaults:
6
+ - _self_
7
+ - inference_experiment: inference_on_user_inputs # Default for design
8
+ - recycle_strategy: thinking_medium
9
+
10
+ # NOTE: The experiment configs can override values specified in this config
11
+
12
+ run_name_: inference
13
+ verbose: false # Set true for routine warnings and detailed diagnostic output.
14
+ hf_ckpt_path: nvidia/RFD4-Proteina # Optional local model bundle or HF repo; overrides ckpt_path/name when set.
15
+ hf_ckpt_revision: a06e2509c7033b010fc43614e9fd840303588502 # Pin a Hub commit for reproducible model/AE loading.
16
+
17
+ out_dir: null # If null, defaults to ./runs/<run_name>. For production, set to an external
18
+ # path (e.g. /shared/runs/experiment_name). SLURM scripts set this
19
+ # automatically. All outputs (samples, metrics) go under this directory.
20
+
21
+ # Current SOTA; generation and regression tooling resolve the immutable checkpoint from here.
22
+ # Regenerate the inference baselines when advancing it; use CLI overrides for one-off checkpoints.
23
+ ckpt_name: rfd4-proteina-flow-release-2026-10-19.ckpt
24
+ ckpt_path: ${oc.env:DATA_PATH}/flow_checkpoints
25
+ adapter_path: null # Local PEFT adapter directory; load onto the exact base/AE pair used for fine-tuning.
26
+
27
+ # Only applies when a pre-oracle design filter is active (i.e. its tag gate matched).
28
+ # Determines whether to delete structure files for designs that fail this filter
29
+ delete_failing_pre_oracle: True
30
+
31
+ autoencoder_ckpt_path: ${oc.env:DATA_PATH}/ae_checkpoints/rfd4-proteina-ae-release-2026-10-19.ckpt
32
+
33
+ num_replicates: 1 # Number of times to replicate each example
34
+
35
+ # Token counts to pad batches to for torch.compile. Multiples of 16; null disables.
36
+ compile_lengths: [128, 160, 192, 224, 256, 288, 320, 352, 384, 448, 512, 640, 768, 896, 1024]
37
+
38
+ # Network settings that may change after training without touching the weights; null keeps the checkpoint's
39
+ # value. Example: nn.torch_compile_mode=reduce-overhead nn.tri_mult_backend=cuequivariance
40
+ nn:
41
+ torch_compile_mode: null
42
+ torch_compile_fullgraph: null
43
+ tri_mult_backend: null
44
+ attention_type: null
45
+
46
+ precision: bf16-mixed
47
+
48
+ batch_size: 1
49
+ subset_to_examples: null # If specified, only run inference on these example indices
50
+ prevalidate_inputs: false # Whether to sanity-check each unique input once prior to generation (fail fast on bad inputs).
51
+ seed: null # Will set a random seed if null
52
+ torch_deterministic: false # Locks cuDNN/cuBLAS to deterministic algorithms; significantly reduces throughput
53
+ skip_existing: true # If true, skip examples whose output structure files already exist in the output directory
54
+ save_chain_of_interest: false # If true, write per-design chain-of-interest CIFs (only needed for foldseek clustering).
55
+ compress_outputs: false # Compress persistent structure outputs, including refolds and CIF trajectories.
56
+ metrics_tags: [] # If specified, determines which metrics are computed during inference
57
+
58
+ search:
59
+ checkpoints: null # Set checkpoint policies to enable beam search.
60
+ save_states: false
61
+ resume_from: null
62
+ resume_seed: null
63
+
64
+ num_workers: 4 # Number of dataloader workers
65
+
66
+ hardware:
67
+ accelerator: gpu
68
+ ngpus_per_node_: 1
69
+ nnodes_: 1
70
+
71
+ # Cluster job settings (used by launch_infer_from_local.sh)
72
+ ncpus_: 24 # Number of CPUs per task
73
+ gen_njobs: 1 # Number of parallel generation jobs per config
74
+ eval_njobs: 1 # Number of parallel evaluation jobs per config
75
+ job_id: null # Set by launch_generate.sh; used to disambiguate per-job output files
76
+
77
+ hydra:
78
+ run:
79
+ dir: ./hydra_outputs
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b549dd61adb2f6c6b5cb95bda13f40f5e31556b4e6c755edc0ca6f011c68fb0
3
+ size 1742660596
openfold3/runtime/of3_af3_refolding_env-cuda12.docker.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4723a0cbc1d83810005c54fe3cf3b881e2023d45412e883879f2fab686ab3620
3
+ size 4406498304
openfold3/runtime/of3_af3_refolding_env-cuda12.docker.tar.sha256 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dcd4fc6523928a57ee858727438576d44556070b56bac1dec27bd530ffff5ec7
3
+ size 106
openfold3/runtime/of3_af3_refolding_env-cuda12.sif ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c2b17e5b9262614844c71ce1ddfa1a03ce3534b367b1e919ea41866027087c71
3
+ size 4000178176
openfold3/runtime/of3_af3_refolding_env-cuda12.sif.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ c2b17e5b9262614844c71ce1ddfa1a03ce3534b367b1e919ea41866027087c71 of3_af3_refolding_env-cuda12.sif
openfold3/runtime/of3_af3_refolding_env-cuda12.sqsh ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6013f17d43d5be70fcad6bb9c674a96803023c418ea0c3dde08f8e1d61ad4719
3
+ size 4000133120
openfold3/runtime/of3_af3_refolding_env-cuda12.sqsh.sha256 ADDED
@@ -0,0 +1 @@
 
 
1
+ 6013f17d43d5be70fcad6bb9c674a96803023c418ea0c3dde08f8e1d61ad4719 of3_af3_refolding_env-cuda12.sqsh
openfold3/runtime/of3_af3_refolding_env-cuda13.docker.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b017dfdf3ea481ad7e56ad8b5fad031f499cb2dc5cafacdeb0a96e84c014fda
3
+ size 3911081472
openfold3/runtime/of3_af3_refolding_env-cuda13.docker.tar.sha256 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3990a966c2ac76ec25818a0d8cb8dd30589c6bc550007ebaf34be2735019d58a
3
+ size 106
openfold3/runtime/of3_af3_refolding_env-cuda13.sif ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bbd35478ed5e95131efa7756d03a2e754f3508e603ad08bb8f8db0bd89484a1
3
+ size 3502800896