You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardexpand all lines: README.md
+1-10
Original file line number
Diff line number
Diff line change
@@ -66,7 +66,7 @@ causal_language_model = LobsterPCLM.load_from_checkpoint(<path to ckpt>)
66
66
```
67
67
3D, cDNA, and dynamic models use the same classes.
68
68
69
-
NOTE: Pre-trained model checkpoints will be included in future releases!
69
+
NOTE: Pre-trained model checkpoints *may* be included in future releases!
70
70
71
71
**Models**
72
72
* LobsterPMLM: masked language model (BERT-style encoder-only architecture)
@@ -96,15 +96,6 @@ model.naturalness(sequences)
96
96
model.likelihood(sequences)
97
97
```
98
98
99
-
## Example Jupyter notebooks
100
-
101
-
### Protein structure prediction
102
-
103
-
see [this notebook](notebooks/01-lobster-fold.ipynb) for an example on using LobsterFold to predict structure from sequence.
104
-
105
-
### Structure-aware sequence embedding with 3D-PPLM
106
-
see [this notebook](notebooks/02-3d-lobster.ipynb) for an example on using the [FoldseekTransform](src/lobster/transforms/_foldseek_transforms.py) and 3D-PPLM to embed a monomer or complex.
107
-
108
99
## Training from scratch
109
100
The entrypoint `lobster_train` is the main driver for training and accepts parameters using Hydra syntax. The available parameters for configuration can be found by running `lobster_train --help` or by looking in the src/lobster/hydra_config directory
Aliquam vitae elit ullamcorper tellus egestas pellentesque. Ut lacus tellus, maximus vel lectus at, placerat pretium mi. Maecenas dignissim tincidunt vestibulum. Sed consequat hendrerit nisl ut maximus.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Proin ullamcorper tellus sed ante aliquam tempus. Etiam porttitor urna feugiat nibh elementum, et tempor dolor mattis. Donec accumsan enim augue, a vulputate nisi sodales sit amet. Proin bibendum ex eget mauris cursus euismod nec et nibh. Maecenas ac gravida ante, nec cursus dui. Vivamus purus nibh, placerat ac purus eget, sagittis vestibulum metus. Sed vestibulum bibendum lectus gravida commodo. Pellentesque auctor leo vitae sagittis suscipit.
151
+
Protein language models (pLMs) are ubiquitous across biological machine learning research, but state-of-the-art models like ESM2 take hundreds of thousands of GPU hours to pre-train on the vast protein universe.
152
+
Resource requirements for scaling up pLMs prevent fundamental investigations into how optimal modeling choices might differ from those used in natural language. Here, we define a “cramming” challenge for pLMs and train
153
+
performant models in 24 hours on a single GPU. By re-examining many aspects of pLM training, we are able to train a 67 million parameter model in a single day that achieves comparable performance on downstream protein fitness
154
+
landscape inference tasks to ESM-3B, a model trained for over 15,000× more GPU hours than ours. We open source our library for training and inference,
155
+
LBSTER: Language models for Biological Sequence Transformation and Evolutionary Representation.
author = {Frey, Nathan C. and Joren, Taylor and Ismail, Aya Abdelsalam and Goodman, Allen and Bonneau, Richard and Cho, Kyunghyun and Gligorijevi{\'c}, Vladimir},
287
+
title = {Cramming Protein Language Model Training in 24 GPU Hours},
This page was built using the <ahref="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> which was adopted from the <ahref="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
295
-
You are free to borrow the of this website, we just ask that you link back to this page in the footer. <br> This website is licensed under a <arel="license" href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
309
+
<br> This website is licensed under a <arel="license" href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
296
310
Commons Attribution-ShareAlike 4.0 International License</a>.
0 commit comments