Loading from Weights & Biases

In [1]:

%load_ext autoreload
%autoreload 2

%load_ext autoreload %autoreload 2

In [2]:

from safe.sample import SAFEDesign

from safe.sample import SAFEDesign

In [3]:

model = SAFEDesign.load_default()

model = SAFEDesign.load_default()

/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(

Upload models to wandb¶

SAFE models can be uploaded to wandb with the upload_to_wandb function. You can define a general "SAFE_WANDB_PROJECT" env variable to save all of your models to that project.

Make sure that you are login into your wandb account:

wandb login --relogin $WANDB_API_KEY

In [4]:

from safe.io import upload_to_wandb

from safe.io import upload_to_wandb

In [5]:

%env WANDB_SILENT=False
%env SAFE_WANDB_PROJECT=safe-models

upload_to_wandb(model.model, model.tokenizer, artifact_name="default-safe-zinc", slicer="BRICS/Partition", aliases=["paper"])

%env WANDB_SILENT=False %env SAFE_WANDB_PROJECT=safe-models upload_to_wandb(model.model, model.tokenizer, artifact_name="default-safe-zinc", slicer="BRICS/Partition", aliases=["paper"])

env: WANDB_SILENT=False
env: SAFE_WANDB_PROJECT=safe-models
[2024-09-10 13:42:46,004] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to mps (auto detect)

W0910 13:42:46.257000 8343047168 torch/distributed/elastic/multiprocessing/redirects.py:28] NOTE: Redirects are currently not supported in Windows or MacOs.
Failed to detect the name of this notebook, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable to enable code saving.
wandb: Currently logged in as: maclandrol (valencelabs). Use `wandb login --relogin` to force relogin

wandb version 0.17.9 is available! To upgrade, please run: $ pip install wandb --upgrade

Tracking run with wandb version 0.16.6

Run data is saved locally in /Users/emmanuel.noutahi/Code/safe/nb/wandb/run-20240910_134247-72wmn5st

Syncing run absurd-disco-1 to Weights & Biases (docs)

View project at https://wandb.ai/valencelabs/safe-models

View run at https://wandb.ai/valencelabs/safe-models/runs/72wmn5st

wandb: Adding directory to artifact (/var/folders/rl/wwcfdj4x0pg293bfqszl970r0000gq/T/tmpy8r3mrzg)... Done. 0.6s

VBox(children=(Label(value='333.221 MB of 333.221 MB uploaded\r'), FloatProgress(value=1.0, max=1.0)))

View run absurd-disco-1 at: https://wandb.ai/valencelabs/safe-models/runs/72wmn5st
View project at: https://wandb.ai/valencelabs/safe-models
Synced 6 W&B file(s), 0 media file(s), 5 artifact file(s) and 0 other file(s)

Find logs at: ./wandb/run-20240910_134247-72wmn5st/logs

Loading models from wandb¶

In [3]:

%env SAFE_MODEL_ROOT=/Users/emmanuel.noutahi/.cache/wandb/safe/
designer = SAFEDesign.load_from_wandb("safe-models/default-safe-zinc")

%env SAFE_MODEL_ROOT=/Users/emmanuel.noutahi/.cache/wandb/safe/ designer = SAFEDesign.load_from_wandb("safe-models/default-safe-zinc")

env: SAFE_MODEL_ROOT=/Users/emmanuel.noutahi/.cache/wandb/safe/

wandb: Downloading large artifact default-safe-zinc:latest, 333.22MB. 5 files... 
wandb:   5 of 5 files downloaded.  
Done. 0:0:0.7

In [4]:

designer.de_novo_generation(10)

designer.de_novo_generation(10)

  0%|          | 0/1 [00:00<?, ?it/s]

/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/transformers/tokenization_utils_base.py:1601: FutureWarning: `clean_up_tokenization_spaces` was not set. It will be set to `True` by default. This behavior will be depracted in transformers v4.45, and will be then set to `False` by default. For more details check this issue: https://github.com/huggingface/transformers/issues/31884
  warnings.warn(
/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/transformers/generation/configuration_utils.py:615: UserWarning: `num_beams` is set to 1. However, `early_stopping` is set to `True` -- this flag is only used in beam-based generation modes. You should set `num_beams>1` or unset `early_stopping`.
  warnings.warn(

Out[4]:

['C[C@]1(C(=O)N2CCC[C@@H](NC(=O)C#CC3CC3)CC2)CCNC1=O',
 'CN(C(=O)CN1CC[NH+](C[C@@H](O)Cn2cc([N+](=O)[O-])cn2)CC1)c1ccccc1',
 'CC[C@@H](C)C[C@@H]([NH3+])C(=O)N(CC)C[C@@H]1CCOC1',
 'Cc1nnc2n1C[C@H](CNC(=O)Nc1cc(Cl)ccc1Cl)CC2',
 'CCc1cccc(CC)c1NC(=O)[C@H](C)OC(=O)CCc1nc2ccccc2o1',
 'Cc1cc(OC[C@H](O)C[NH2+]C[C@@H]2C[C@H](O)CN2Cc2ccccc2)ccc1F',
 'Cc1c(Cl)cccc1N=C(O)CN=C(O)COC(=O)c1csc(-c2ccccc2)n1',
 'CCc1nc(CCNC(=O)N[C@@H]2CCc3nnnn3CC2)cs1',
 'C[C@@]1(C(=O)N[C@H]2CCCCCN(C(=O)c3cc(C4CC4)no3)C2)C=CCC1',
 'Cc1cc(-c2cc(-c3cnn(C)c3)c3c(N)ncnc3n2)ccc1F']