Ray Gao
Hi, this is a very interesting work! I was just curious if you had some insights on why the cauchy filter worked for the attentive sampler over the latent space...
This addresses https://github.com/facebookresearch/fairchem/issues/1189: a bug where when activation checkpointing was turned on but MoLE was unmerged (`default` inference mode). This happens when the system is large enough that requires activation...
* This allows GPU graph gen to be partitioned across ranks. Each rank is responsible for generating it's own graph partition by generating the graph with it's own target atoms...
### Python version 3.11 ### fairchem-core version 2.11 ### pytorch version 2.8.0 ### cuda version 12.1 ### Operating system version _No response_ ### Minimal example ```Python ``` ### Current behavior...
### 🐛 Describe the bug can we fix the annoying pkg_resource warning here? :) ../lib/python3.12/site-packages/torchtnt/utils/version.py:12: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for...