Nathan Lambert

Results 60 issues of Nathan Lambert

The score-based generative models sometimes us a Langevin Dynamics step to de-noise samples; see [NCSN](https://arxiv.org/abs/2006.09011). In the diffusion model for generating molecule conformations, [GeoDiff](https://github.com/MinkaiXu/GeoDiff), this step function had the best...

good first issue

Added a new model file to add functionality for [this paper](https://github.com/MinkaiXu/GeoDiff) that does molecule generation via diffusion. Pretrained models available are for two tasks, `drugs` and `qm9`: ```python model1 =...

Creating the PR to add back in the RL code when ready. This contains the model to run inference for Janner et. al's [Diffuser](https://github.com/jannerm/diffuser), with the accompanying [colab](https://colab.research.google.com/drive/1TmBmlYeKUZSkUZoJqfBmaicVTKx6nN1R?usp=sharing). _**work in...

We're discussing where certain properties should be saved in this repository. For example, [this `predict_epsilon`](https://github.com/huggingface/diffusers/blob/320506c75ad7d40c7d9fc6db5a45955906c5f729/src/diffusers/schedulers/scheduling_ddpm.py#L125) property is used to distinguish if a model is trained to directly reconstruct the original...

TODO for this WIP PR: - [x] New PID based / linear feedback agent(s) - [ ] Make PID accept vector inputs - [x] Training example - [ ] Migrate...

CLA Signed

# 🚀 Feature Request When training non delta-state models, the outputs of dynamics models can take large values (way outside a unit Gaussian). In the past I have tried using...

enhancement

# 🚀 Feature Request A configuration option to save all data for every trial. ## Motivation When debugging it can be very useful to see the intricate details of how...

enhancement

The function found [here](https://github.com/MinkaiXu/GeoDiff/blob/c6f26dc250308bff8923a19884e601e0bb0f975a/models/encoder/edge.py#L65) uses the full variable `config` which is not passed in for the type `gaussian`. I believe it should be `cfg`, but I guess it is never...

# 🚀 Feature Request Add functionality to upload dynamics models /policies to the HF hug at end of training or during training for sharing / fine-tuning. This would like like...

enhancement