Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -80,7 +80,7 @@ Encoders per modality:
|
|
| 80 |
- Fingerprint Transformer (PolyFusion/Transformer.py)
|
| 81 |
|
| 82 |
- Projects each modality into a shared, unit-normalized latent space
|
| 83 |
-
- Uses contrastive alignment where a fused structural anchor (PSMILES + 2D + 3D) is aligned with a fingerprint target (PolyFusion/CL.py)
|
| 84 |
|
| 85 |
### Downstream Tasks
|
| 86 |
|
|
@@ -89,8 +89,7 @@ Encoders per modality:
|
|
| 89 |
- Returns predictions in original units (with standardization handled internally)
|
| 90 |
|
| 91 |
#### 4) Inverse design / polymer generation (targets → candidates) (Downstream Tasks/Polymer_Generation.py)
|
| 92 |
-
- Property-conditioned candidate generation using PolyFusion embeddings as the conditioning interface
|
| 93 |
-
- Supports optional seeding/biasing (e.g., start from a reference polymer family)
|
| 94 |
- Produces candidate lists suitable for generate → filter → validate workflows
|
| 95 |
|
| 96 |
### PolyAgent
|
|
@@ -116,8 +115,6 @@ This repository is configured as a Gradio Space via the YAML header at the top o
|
|
| 116 |
|
| 117 |
Entry point: app_file: PolyAgent/gradio_interface.py
|
| 118 |
|
| 119 |
-
Space URL: kaurm43/PolyFusionAgent
|
| 120 |
-
|
| 121 |
---
|
| 122 |
|
| 123 |
## Model weights and artifacts
|
|
@@ -131,15 +128,13 @@ POLYFUSION_WEIGHTS_REPO=kaurm43/polyfusionagent-weights
|
|
| 131 |
|
| 132 |
POLYFUSION_WEIGHTS_REPO_TYPE=model
|
| 133 |
|
| 134 |
-
Weights repo
|
| 135 |
|
| 136 |
-
Weights files: weights/tree/main
|
| 137 |
|
| 138 |
### Override via environment variables (local or Space secrets)
|
| 139 |
POLYFUSION_WEIGHTS_REPO=your-org/your-weights-repo
|
| 140 |
POLYFUSION_WEIGHTS_REPO_TYPE=model
|
| 141 |
POLYFUSION_WEIGHTS_DIR=/path/to/cache
|
| 142 |
-
HF_TOKEN=... # only if private
|
| 143 |
|
| 144 |
### Expected weights layout (inside the weights repo)
|
| 145 |
The orchestrator expects these folders/files:
|
|
@@ -192,16 +187,7 @@ python PolyAgent/gradio_interface.py
|
|
| 192 |
---
|
| 193 |
|
| 194 |
## Configuration (optional)
|
| 195 |
-
|
| 196 |
-
### Common environment variables
|
| 197 |
```bash
|
| 198 |
-
# Weights / cache
|
| 199 |
-
POLYFUSION_WEIGHTS_REPO=kaurm43/polyfusionagent-weights
|
| 200 |
-
POLYFUSION_WEIGHTS_REPO_TYPE=model
|
| 201 |
-
POLYFUSION_WEIGHTS_DIR=.cache/polyfusion_weights
|
| 202 |
-
|
| 203 |
-
# Retrieval DB path
|
| 204 |
-
CHROMA_DB_PATH=chroma_polymer_db_big
|
| 205 |
|
| 206 |
# - If someone forks/runs locally and needs these APIs, they could create their OWN keys and set them as
|
| 207 |
# environment variables (or add them in their own Space Settings → Secrets).
|
|
|
|
| 80 |
- Fingerprint Transformer (PolyFusion/Transformer.py)
|
| 81 |
|
| 82 |
- Projects each modality into a shared, unit-normalized latent space
|
| 83 |
+
- Uses contrastive alignment where a fused structural anchor (PSMILES + 2D graph + 3D geometry) is aligned with a fingerprint target (PolyFusion/CL.py)
|
| 84 |
|
| 85 |
### Downstream Tasks
|
| 86 |
|
|
|
|
| 89 |
- Returns predictions in original units (with standardization handled internally)
|
| 90 |
|
| 91 |
#### 4) Inverse design / polymer generation (targets → candidates) (Downstream Tasks/Polymer_Generation.py)
|
| 92 |
+
- Property-conditioned candidate generation using PolyFusion embeddings as the conditioning interface for a sequence-to-sequence generator (SELFIES-based Encoder Decoder)
|
|
|
|
| 93 |
- Produces candidate lists suitable for generate → filter → validate workflows
|
| 94 |
|
| 95 |
### PolyAgent
|
|
|
|
| 115 |
|
| 116 |
Entry point: app_file: PolyAgent/gradio_interface.py
|
| 117 |
|
|
|
|
|
|
|
| 118 |
---
|
| 119 |
|
| 120 |
## Model weights and artifacts
|
|
|
|
| 128 |
|
| 129 |
POLYFUSION_WEIGHTS_REPO_TYPE=model
|
| 130 |
|
| 131 |
+
- **Weights repo:** [kaurm43/polyfusionagent-weights](https://huggingface.co/kaurm43/polyfusionagent-weights)
|
| 132 |
|
|
|
|
| 133 |
|
| 134 |
### Override via environment variables (local or Space secrets)
|
| 135 |
POLYFUSION_WEIGHTS_REPO=your-org/your-weights-repo
|
| 136 |
POLYFUSION_WEIGHTS_REPO_TYPE=model
|
| 137 |
POLYFUSION_WEIGHTS_DIR=/path/to/cache
|
|
|
|
| 138 |
|
| 139 |
### Expected weights layout (inside the weights repo)
|
| 140 |
The orchestrator expects these folders/files:
|
|
|
|
| 187 |
---
|
| 188 |
|
| 189 |
## Configuration (optional)
|
|
|
|
|
|
|
| 190 |
```bash
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 191 |
|
| 192 |
# - If someone forks/runs locally and needs these APIs, they could create their OWN keys and set them as
|
| 193 |
# environment variables (or add them in their own Space Settings → Secrets).
|