Stars
The offical Implementation of "Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model"
Real-Time VLAs via Future-state-aware Asynchronous Inference.
Progressive CUDA GEMM suite covering kernels 01–09: naive & anti-patterns through shared/reg tiling, float4/vector, autotuned tiles, WMMA, and cp.async double-buffered pipelines, plus cuBLAS refs. …
MinIO is a high-performance, S3 compatible object store, open sourced under GNU AGPLv3 license.
VLA-0: Building State-of-the-Art VLAs with Zero Modification
Complete solutions to the Programming Massively Parallel Processors Edition 4
Faster Pi0-FAST inference time; DCT early stopping trick implementation for FAST (Frequency-Space Action Sequence Tokenization)
A Modular Toolkit for Robot Kinematic Optimization
Realtime & high-frequency control interfaces for various robot arms including bi-manual I2RT YAM, Franka Panda, with manual tele-operation control or autonomous policy control
SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning
Parser for GPMF™ formatted telemetry data used within GoPro® cameras.
A Scalable and Hardware-Independent Universal Manipulation Interface
Minimal reproduction of DeepSeek R1-Zero
Unified high-performance Python client for object and file stores.
Math Olympiad Agent solves IMO-level problems with a self-verification pipeline, inspired by Huang & Yang’s 2025 paper. Built with LangGraph and A2A protocol for agent collaboration, it generates r…
Fast, automatic, accurate camera calibration. A simplified and expanded version of EasyHec
A Datacenter Scale Distributed Inference Serving Framework
Pretraining data reconstruction scripts for Apertus