ZhiyuanChen commited on
Commit
89b64d0
·
verified ·
1 Parent(s): 72b400e

Publish OFoldX pipeline artifacts

Browse files
README.md CHANGED
@@ -6,11 +6,11 @@ tags:
6
  - "protein"
7
  - "rna"
8
  - "dna"
9
- - "model"
10
  - "boltzgen-adherence"
11
  - "design-generation"
12
  - "protein-design"
13
- artifact_kind: "model"
14
  repo_id: "oteam/boltzgen-adherence"
15
  license: "mit"
16
  base_model: "boltz-community/boltz-2"
@@ -38,11 +38,11 @@ widget:
38
 
39
  # boltzgen-adherence
40
 
41
- OFoldX `model` artifact for biomolecular design generation, using the `boltzgen-adherence` architecture.
42
 
43
  ## Disclaimer
44
 
45
- This model card was generated by the OFoldX team for an OFoldX `model` artifact.
46
  The upstream model authors did not write this card unless explicitly stated otherwise.
47
 
48
  OFoldX is pre-alpha research software. Check the source checkpoint, upstream release, and local validation
@@ -67,12 +67,10 @@ Converted BoltzGen adherence generator checkpoint for controllable binder and pr
67
  | Field | Value |
68
  | ----- | ----- |
69
  | Repository | `oteam/boltzgen-adherence` |
70
- | Artifact Kind | `model` |
71
  | Task | `design_generation` |
72
  | Architecture | `boltzgen-adherence` |
73
- | Entrypoint | `ofoldx.models.boltzgen.model.BoltzGenDesignGenerator` |
74
- | Config | `config.json` |
75
- | Weights | `model.safetensors` |
76
  | Source Checkpoint | `boltzgen1_adherence.ckpt` |
77
 
78
  > [!NOTE]
@@ -84,7 +82,7 @@ Converted BoltzGen adherence generator checkpoint for controllable binder and pr
84
  - **Upstream paper**: [BoltzGen: Toward Universal Binder Design](https://doi.org/10.1101/2025.11.20.689494)
85
  - **Upstream repository**: [BoltzGen](https://github.com/HannesStark/boltzgen)
86
  - **Source checkpoint release**: [https://huggingface.co/boltzgen/boltzgen-1](https://huggingface.co/boltzgen/boltzgen-1)
87
- - **Code**: [`ofoldx/models/boltzgen/model.py`](https://github.com/OTeam-AI4S/OFoldX/tree/main/ofoldx/models/boltzgen/model.py)
88
  - **Project repository**: [https://github.com/OTeam-AI4S/OFoldX](https://github.com/OTeam-AI4S/OFoldX)
89
  - **Issues**: [https://github.com/OTeam-AI4S/OFoldX/issues](https://github.com/OTeam-AI4S/OFoldX/issues)
90
 
@@ -96,19 +94,14 @@ The artifact depends on the [`ofoldx`](https://github.com/OTeam-AI4S/OFoldX) lib
96
  pip install ofoldx
97
  ```
98
 
99
- ### Direct Use
100
 
101
- Load the artifact from `oteam/boltzgen-adherence` using the OFoldX `Auto*` interface:
102
 
103
  ```python
104
  from ofoldx.pipelines import Pipeline
105
 
106
  pipeline = Pipeline.from_pretrained("oteam/boltzgen-adherence")
107
- output = pipeline(...)
108
-
109
- from ofoldx import AutoModel
110
-
111
- model = AutoModel.from_pretrained("oteam/boltzgen-adherence")
112
  ```
113
 
114
  When a matching processor is available, load it with `AutoProcessor.from_pretrained(...)` and pass the
@@ -117,7 +110,7 @@ processed batch to the model.
117
  ### Interface
118
 
119
  - **Task**: `design_generation`
120
- - **Artifact kind**: `model`
121
  - **Architecture**: `boltzgen-adherence`
122
  - **Runtime files**: `manifest.json`, `config.json`, and `model.safetensors` when present
123
 
@@ -165,5 +158,5 @@ Please use [OFoldX GitHub issues](https://github.com/OTeam-AI4S/OFoldX/issues) f
165
 
166
  ## License
167
 
168
- The OFoldX project license is not yet finalized.
169
  The source checkpoint is associated with the upstream license noted above: MIT for upstream BoltzGen code. Review both OFoldX and upstream terms before redistribution or production use.
 
6
  - "protein"
7
  - "rna"
8
  - "dna"
9
+ - "pipeline"
10
  - "boltzgen-adherence"
11
  - "design-generation"
12
  - "protein-design"
13
+ artifact_kind: "pipeline"
14
  repo_id: "oteam/boltzgen-adherence"
15
  license: "mit"
16
  base_model: "boltz-community/boltz-2"
 
38
 
39
  # boltzgen-adherence
40
 
41
+ OFoldX `pipeline` artifact for biomolecular design generation, using the `boltzgen-adherence` architecture.
42
 
43
  ## Disclaimer
44
 
45
+ This model card was generated by the OFoldX team for an OFoldX `pipeline` artifact.
46
  The upstream model authors did not write this card unless explicitly stated otherwise.
47
 
48
  OFoldX is pre-alpha research software. Check the source checkpoint, upstream release, and local validation
 
67
  | Field | Value |
68
  | ----- | ----- |
69
  | Repository | `oteam/boltzgen-adherence` |
70
+ | Artifact Kind | `pipeline` |
71
  | Task | `design_generation` |
72
  | Architecture | `boltzgen-adherence` |
73
+ | Entrypoint | `ofoldx.pipelines.design.DesignPipeline` |
 
 
74
  | Source Checkpoint | `boltzgen1_adherence.ckpt` |
75
 
76
  > [!NOTE]
 
82
  - **Upstream paper**: [BoltzGen: Toward Universal Binder Design](https://doi.org/10.1101/2025.11.20.689494)
83
  - **Upstream repository**: [BoltzGen](https://github.com/HannesStark/boltzgen)
84
  - **Source checkpoint release**: [https://huggingface.co/boltzgen/boltzgen-1](https://huggingface.co/boltzgen/boltzgen-1)
85
+ - **Code**: [`ofoldx/pipelines/design.py`](https://github.com/OTeam-AI4S/OFoldX/tree/main/ofoldx/pipelines/design.py)
86
  - **Project repository**: [https://github.com/OTeam-AI4S/OFoldX](https://github.com/OTeam-AI4S/OFoldX)
87
  - **Issues**: [https://github.com/OTeam-AI4S/OFoldX/issues](https://github.com/OTeam-AI4S/OFoldX/issues)
88
 
 
94
  pip install ofoldx
95
  ```
96
 
97
+ ### Pipeline Usage
98
 
99
+ Load the artifact from `oteam/boltzgen-adherence` with the OFoldX task pipeline. Use `AutoModel` or `AutoProcessor` only when you need lower-level control:
100
 
101
  ```python
102
  from ofoldx.pipelines import Pipeline
103
 
104
  pipeline = Pipeline.from_pretrained("oteam/boltzgen-adherence")
 
 
 
 
 
105
  ```
106
 
107
  When a matching processor is available, load it with `AutoProcessor.from_pretrained(...)` and pass the
 
110
  ### Interface
111
 
112
  - **Task**: `design_generation`
113
+ - **Artifact kind**: `pipeline`
114
  - **Architecture**: `boltzgen-adherence`
115
  - **Runtime files**: `manifest.json`, `config.json`, and `model.safetensors` when present
116
 
 
158
 
159
  ## License
160
 
161
+ The Hub `license` metadata, when present, reflects the source checkpoint or upstream project license. The OFoldX project license is not yet finalized.
162
  The source checkpoint is associated with the upstream license noted above: MIT for upstream BoltzGen code. Review both OFoldX and upstream terms before redistribution or production use.
generator/README.md ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: "ofoldx"
3
+ tags:
4
+ - "biology"
5
+ - "biomolecular-design"
6
+ - "protein"
7
+ - "rna"
8
+ - "dna"
9
+ - "model"
10
+ - "boltzgen-adherence"
11
+ - "design-generation"
12
+ - "protein-design"
13
+ artifact_kind: "model"
14
+ repo_id: "oteam/boltzgen-adherence"
15
+ license: "mit"
16
+ base_model: "boltz-community/boltz-2"
17
+ datasets:
18
+ - "boltzgen/boltzgen1_train"
19
+ - "boltzgen/inference-data"
20
+ pipeline_tag: "other"
21
+ task: "design_generation"
22
+ model-index:
23
+ - name: "boltzgen-adherence"
24
+ results:
25
+ []
26
+ widget:
27
+ - pipeline_tag: "other"
28
+ task: "design_generation"
29
+ example_title: "Backbone sequence design"
30
+ text: "input_structure: backbone.cif\ndesign_chains: A"
31
+ input_format: "structure_path"
32
+ - pipeline_tag: "other"
33
+ task: "design_generation"
34
+ example_title: "Binder design"
35
+ text: "target_structure: target.cif\ntarget_chains: A\ndesign_chains: B"
36
+ input_format: "structure_path"
37
+ ---
38
+
39
+ # boltzgen-adherence
40
+
41
+ OFoldX `model` artifact for biomolecular design generation, using the `boltzgen-adherence` architecture.
42
+
43
+ ## Disclaimer
44
+
45
+ This model card was generated by the OFoldX team for an OFoldX `model` artifact.
46
+ The upstream model authors did not write this card unless explicitly stated otherwise.
47
+
48
+ OFoldX is pre-alpha research software. Check the source checkpoint, upstream release, and local validation
49
+ before using the artifact for scientific or operational decisions.
50
+
51
+ ## Model Details
52
+
53
+ BoltzGen design generator variant optimized for adherence to conditioning signals.
54
+
55
+ Converted BoltzGen adherence generator checkpoint for controllable binder and protein design.
56
+
57
+ ### Model Provenance
58
+
59
+ - **Upstream Project**: BoltzGen
60
+ - **Source Checkpoint**: `boltzgen1_adherence.ckpt`
61
+ - **Source Release**: [https://huggingface.co/boltzgen/boltzgen-1](https://huggingface.co/boltzgen/boltzgen-1)
62
+ - **Primary Paper**: [BoltzGen: Toward Universal Binder Design](https://doi.org/10.1101/2025.11.20.689494)
63
+ - **Upstream License**: MIT for upstream BoltzGen code
64
+
65
+ ### Model Specification
66
+
67
+ | Field | Value |
68
+ | ----- | ----- |
69
+ | Repository | `oteam/boltzgen-adherence` |
70
+ | Artifact Kind | `model` |
71
+ | Task | `design_generation` |
72
+ | Architecture | `boltzgen-adherence` |
73
+ | Entrypoint | `ofoldx.models.boltzgen.model.BoltzGenDesignGenerator` |
74
+ | Config | `config.json` |
75
+ | Weights | `model.safetensors` |
76
+ | Source Checkpoint | `boltzgen1_adherence.ckpt` |
77
+
78
+ > [!NOTE]
79
+ > Source checkpoint: `boltzgen1_adherence.ckpt`.
80
+
81
+ ### Links
82
+
83
+ - **Hub repository**: [oteam/boltzgen-adherence](https://huggingface.co/oteam/boltzgen-adherence)
84
+ - **Upstream paper**: [BoltzGen: Toward Universal Binder Design](https://doi.org/10.1101/2025.11.20.689494)
85
+ - **Upstream repository**: [BoltzGen](https://github.com/HannesStark/boltzgen)
86
+ - **Source checkpoint release**: [https://huggingface.co/boltzgen/boltzgen-1](https://huggingface.co/boltzgen/boltzgen-1)
87
+ - **Code**: [`ofoldx/models/boltzgen/model.py`](https://github.com/OTeam-AI4S/OFoldX/tree/main/ofoldx/models/boltzgen/model.py)
88
+ - **Project repository**: [https://github.com/OTeam-AI4S/OFoldX](https://github.com/OTeam-AI4S/OFoldX)
89
+ - **Issues**: [https://github.com/OTeam-AI4S/OFoldX/issues](https://github.com/OTeam-AI4S/OFoldX/issues)
90
+
91
+ ## Usage
92
+
93
+ The artifact depends on the [`ofoldx`](https://github.com/OTeam-AI4S/OFoldX) library. Install it with pip:
94
+
95
+ ```bash
96
+ pip install ofoldx
97
+ ```
98
+
99
+ ### Pipeline Usage
100
+
101
+ Load the artifact from `oteam/boltzgen-adherence` with the OFoldX task pipeline. Use `AutoModel` or `AutoProcessor` only when you need lower-level control:
102
+
103
+ ```python
104
+ from ofoldx.pipelines import Pipeline
105
+
106
+ pipeline = Pipeline.from_pretrained("oteam/boltzgen-adherence")
107
+ output = pipeline(...)
108
+
109
+ from ofoldx import AutoModel
110
+
111
+ model = AutoModel.from_pretrained("oteam/boltzgen-adherence")
112
+ ```
113
+
114
+ When a matching processor is available, load it with `AutoProcessor.from_pretrained(...)` and pass the
115
+ processed batch to the model.
116
+
117
+ ### Interface
118
+
119
+ - **Task**: `design_generation`
120
+ - **Artifact kind**: `model`
121
+ - **Architecture**: `boltzgen-adherence`
122
+ - **Runtime files**: `manifest.json`, `config.json`, and `model.safetensors` when present
123
+
124
+ ## Training Details
125
+
126
+ OFoldX did not train these weights. This repository contains a converted checkpoint and OFoldX runtime
127
+ metadata for loading it.
128
+
129
+ ### Training Data
130
+
131
+ BoltzGen builds on the Boltz-2 data pipeline. The public training release includes `targets.zip` and `msa.zip` in the `boltzgen/boltzgen1_train` dataset and molecule dictionaries in `boltzgen/inference-data`; the full large-model recipe may also use additional distillation data.
132
+
133
+ ### Training Procedure
134
+
135
+ Upstream BoltzGen trains a diffusion objective over randomly cropped biomolecular structures with randomly selected design and conditioning regions. OFoldX converts the released adherence checkpoint; it does not run BoltzGen training.
136
+
137
+ ## Evaluation
138
+
139
+ OFoldX conversion reports and contract tests validate artifact structure and checkpoint loading. Task-level
140
+ scientific evaluation should be checked against the corresponding upstream model release or paper.
141
+
142
+ ## Limitations
143
+
144
+ - This artifact is distributed for research use.
145
+ - Inputs must match the model-specific processor and expected biomolecular representation.
146
+ - OFoldX is pre-alpha, so APIs and artifact metadata may still change before a stable release.
147
+
148
+ ## Citation
149
+
150
+ Please cite the upstream BoltzGen work for the source checkpoint. If OFoldX supports your work, please also cite or link the OFoldX project repository.
151
+
152
+ ```bibtex
153
+ @article{stark2025boltzgen,
154
+ author = {Stark, Hannes and Faltings, Felix and Choi, MinGyu and Xie, Yuxin and Hur, Eunsu and O'Donnell, Timothy John and Bushuiev, Anton and Ucar, Talip and Passaro, Saro and Mao, Weian and others},
155
+ title = {BoltzGen: Toward Universal Binder Design},
156
+ year = {2025},
157
+ doi = {10.1101/2025.11.20.689494},
158
+ journal = {bioRxiv}
159
+ }
160
+ ```
161
+
162
+ ## Contact
163
+
164
+ Please use [OFoldX GitHub issues](https://github.com/OTeam-AI4S/OFoldX/issues) for questions or comments about this model card.
165
+
166
+ ## License
167
+
168
+ The Hub `license` metadata, when present, reflects the source checkpoint or upstream project license. The OFoldX project license is not yet finalized.
169
+ The source checkpoint is associated with the upstream license noted above: MIT for upstream BoltzGen code. Review both OFoldX and upstream terms before redistribution or production use.
config.json → generator/config.json RENAMED
File without changes
generator/manifest.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "biomolecular_pretrained",
3
+ "format_version": 1,
4
+ "kind": "model",
5
+ "producer": {
6
+ "package": "ofoldx",
7
+ "version": "0.1.dev1+g9a78c7e44.d20260609"
8
+ },
9
+ "entrypoint": "ofoldx.models.boltzgen.model.BoltzGenDesignGenerator",
10
+ "architecture": "boltzgen_adherence",
11
+ "task": "design_generation",
12
+ "files": {
13
+ "config": "config.json"
14
+ },
15
+ "weights": {
16
+ "default": "model.safetensors"
17
+ }
18
+ }
model.safetensors → generator/model.safetensors RENAMED
File without changes
manifest.json CHANGED
@@ -1,18 +1,22 @@
1
  {
2
  "format": "biomolecular_pretrained",
3
  "format_version": 1,
4
- "kind": "model",
5
  "producer": {
6
  "package": "ofoldx",
7
  "version": "0.1.dev1+g9a78c7e44.d20260609"
8
  },
9
- "entrypoint": "ofoldx.models.boltzgen.model.BoltzGenDesignGenerator",
10
  "architecture": "boltzgen_adherence",
11
  "task": "design_generation",
12
- "files": {
13
- "config": "config.json"
14
- },
15
- "weights": {
16
- "default": "model.safetensors"
 
 
 
 
17
  }
18
  }
 
1
  {
2
  "format": "biomolecular_pretrained",
3
  "format_version": 1,
4
+ "kind": "pipeline",
5
  "producer": {
6
  "package": "ofoldx",
7
  "version": "0.1.dev1+g9a78c7e44.d20260609"
8
  },
9
+ "entrypoint": "ofoldx.pipelines.design.DesignPipeline",
10
  "architecture": "boltzgen_adherence",
11
  "task": "design_generation",
12
+ "components": {
13
+ "generator": {
14
+ "kind": "component",
15
+ "path": "generator"
16
+ },
17
+ "processor": {
18
+ "kind": "component",
19
+ "path": "processor"
20
+ }
21
  }
22
  }
processor/manifest.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "biomolecular_pretrained",
3
+ "format_version": 1,
4
+ "kind": "processor",
5
+ "producer": {
6
+ "package": "ofoldx",
7
+ "version": "0.1.dev1+g9a78c7e44.d20260609"
8
+ },
9
+ "entrypoint": "ofoldx.models.boltzgen.processor.BoltzGenProcessor",
10
+ "files": {
11
+ "config": "processor_config.json"
12
+ }
13
+ }
processor/processor_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "reference_conformer_augment": false,
3
+ "augment_seed": null,
4
+ "materialize_missing_atoms": false,
5
+ "method": "x-ray diffraction"
6
+ }