exalearn / fast-finetuned-forcefields / 5867624077
90%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 23
Run time 2s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
5867624077

push

github

web-flow 
Fine-tune models using MP2 data (#38)

* Switch to using Perlmutter, small basis set pretasks

* Add notebooks to start fine-tuning on MP2

* Go back to dft guesses, update plots

* Add notebooks for comparing different pre-trained models

* Add ability to run MPI NWChem tasks

* Add NWChem to the build environment

* Fix the directory in which a calculation runs

Will enable restarts because the same inputs will be directed to the
same directory

* Update for Perlmutter

* Finish perlmutter port, add restarting, multiple jobs

* Update the comparison notebooks

* Allow sampling to be placed on GPU nodes

We cannot run single-node sampling jobs on the same executor
as multi-node NWChem runs

* Remove vestiges of older Colmena

1. Colmena records proxy stats for us
2. It also measures the runtimes

* Fix a few backwards conditional statements

* Change how we choose block sizes

* Finish the MP2//AVTZ initial set

* Removed outdated environment file

* Add environment for Peralmutter GPU

* Reset the command after execution

* Add multi-node support to run.py

* Add redis configuration file

* Fix Parsl configuration

1. Ensure srun gives Parsl workers access to GPUs
2. Properly active environment (source not conda)

* Fix node count logic, argparse choices

* Regular scaling works fine

* Fix lockup where simulation tasks stop processing

They were waiting to allocate more resources to sampling,
which won't become available until the current round of sampling
finishes.

* Run configuration used on perlmutter

* Test training with only MP2 data

982 of 1097 relevant lines covered (89.52%)

0.9 hits per line

Jobs
ID Job ID Ran Files Coverage
1 5867624077.1 23
89.52
Source Files on build 5867624077