Commit
•
bf7503b
1
Parent(s):
b97ce95
readme changes
Browse files
README.md
CHANGED
|
@@ -1,7 +1,6 @@
|
|
| 1 |
---
|
| 2 |
tags:
|
| 3 |
- chemistry
|
| 4 |
-
- medical
|
| 5 |
widget:
|
| 6 |
- text: <LIGAND>
|
| 7 |
example_title: Generate molecule
|
|
@@ -21,8 +20,8 @@ The model was pretrained on the [Uni-Mol](https://github.com/deepmodeling/Uni-Mo
|
|
| 21 |
the distribution of the GEOM-DRUGS datasets.
|
| 22 |
We also expose pretrained and finetuned models:
|
| 23 |
|
| 24 |
-
- For the pretrained model, visit [huggingface.co/insilicomedicine/
|
| 25 |
-
-
|
| 26 |
|
| 27 |
|
| 28 |
## Unconditional generation
|
|
@@ -33,15 +32,14 @@ sampling molecules from the model. It only depends on
|
|
| 33 |
and it's not meant to reproduce the sampling speed reported
|
| 34 |
in the paper (e.g. it does not use flash-attention, mixed precision,
|
| 35 |
and large batch sampling).
|
| 36 |
-
To reproduce sampling speed, please use the code from our repository:
|
| 37 |
-
https://github.com/insilicomedicine/bindgpt
|
| 38 |
|
| 39 |
```python
|
| 40 |
# Download model from Hugginface:
|
| 41 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 42 |
|
| 43 |
-
tokenizer = AutoTokenizer.from_pretrained("
|
| 44 |
-
model = AutoModelForCausalLM.from_pretrained("
|
| 45 |
|
| 46 |
# Generate 10 tokenized molecules without condition
|
| 47 |
NUM_SAMPLES = 10
|
|
|
|
| 1 |
---
|
| 2 |
tags:
|
| 3 |
- chemistry
|
|
|
|
| 4 |
widget:
|
| 5 |
- text: <LIGAND>
|
| 6 |
example_title: Generate molecule
|
|
|
|
| 20 |
the distribution of the GEOM-DRUGS datasets.
|
| 21 |
We also expose pretrained and finetuned models:
|
| 22 |
|
| 23 |
+
- For the pretrained model, visit [huggingface.co/insilicomedicine/bindgpt_pretrained](https://huggingface.co/insilicomedicine/bindgpt_pretrained)
|
| 24 |
+
- The model finetuned with Reinforcement Learning on CrossDocked is coming soon
|
| 25 |
|
| 26 |
|
| 27 |
## Unconditional generation
|
|
|
|
| 32 |
and it's not meant to reproduce the sampling speed reported
|
| 33 |
in the paper (e.g. it does not use flash-attention, mixed precision,
|
| 34 |
and large batch sampling).
|
| 35 |
+
To reproduce sampling speed, please use the code from our repository: (code coming soon)
|
|
|
|
| 36 |
|
| 37 |
```python
|
| 38 |
# Download model from Hugginface:
|
| 39 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 40 |
|
| 41 |
+
tokenizer = AutoTokenizer.from_pretrained("insilicomedicine/bindgpt_finetuned")
|
| 42 |
+
model = AutoModelForCausalLM.from_pretrained("insilicomedicine/bindgpt_finetuned").cuda()
|
| 43 |
|
| 44 |
# Generate 10 tokenized molecules without condition
|
| 45 |
NUM_SAMPLES = 10
|