
Finds and fixes wrong gradients in differentiable NVIDIA Warp programs using
- Open source
- 4K GitHub stars
- Developer tools

Finds and fixes wrong gradients in differentiable NVIDIA Warp programs using

Checks whether an existing code path is a credible NVIDIA Warp acceleration candidate

Helps agents find and recommend NVIDIA skills from the live catalog before general answers

Guides an agent through GPU DataFrame work with NVIDIA cuDF and dask-cuDF

Guides an agent through customizing NVIDIA's Nemotron Voice Agent for healthcare intake

Calibrates camera extrinsics from live RTSP streams via the AutoMagicCalib REST API

Calibrates multi-camera MP4 recordings through the AutoMagicCalib REST API

Launches the AutoMagicCalib microservice and web UI from NGC images with Docker Compose

Predicts biomolecular structures and ligand binding affinity via Boltz2 NIM

Validates coding-sequence CSVs and extracts frozen CodonFM embeddings for mRNA modeling

Fine-tunes public CodonFM Encodon checkpoints on labeled coding-sequence data

Runs masked-codon variant scoring with the public CodonFM Encodon models

Sets up the public CodonFM repo, dev container and Encodon checkpoint downloads

Scores existing PDB design files with refolding and interface or motif metrics

Runs and ranks cartesian-product parameter sweeps for Proteina-Complexa protein design

Adds and checks Proteina-Complexa design target definitions in YAML registries

Guides building equivariant tensor products and segmented polynomials with cuEquivariance

Guides an agent through writing and debugging cuEquivariance JAX polynomials and Flax NNX

Guides building and debugging cuEquivariance PyTorch tensor products and equivariant GNNs

Predicts protein-ligand binding poses with DiffDock through NVIDIA NIM, hosted or local

Chains GenMol, DiffDock and Boltz2 NIMs to generate, dock and score drug candidates

Generates DNA sequences with NVIDIA's Evo 2 model and extracts local layer outputs

Generates novel drug-like molecules with the GenMol NIM microservice

Adds optional NVIDIA Parabricks GPU steps to genomics pipelines behind a runtime toggle

Converts a GROVER checkpoint into a hybrid cMIM model and continues pretraining it

Continues KERMT molecular pretraining on a custom SMILES corpus inside a GPU container

Extracts per-molecule embeddings from KERMT checkpoints in a containerized GPU run

Finetunes a pretrained KERMT molecular encoder on your labeled SMILES CSV in Docker

Runs predictions with a finetuned KERMT checkpoint on a SMILES-only CSV

Checks progress of a detached KERMT training run: container state, logs and metrics