mjbatch

Jolt Physics is an open-source, multicore-friendly C++ rigid-body physics and collision-detection library for games and VR applications. It simulates rigid bodies with continuous collision detection across shapes such as spheres, boxes, capsules, convex hulls, meshes, and terrain; supports constraints including fixed, hinge, slider, gear, pulley, spline, and six-degrees-of-freedom joints; and provides ray casting, shape queries, broad-phase detection, sensors, vehicles, ragdolls, soft bodies, and buoyancy. Its design supports concurrent physics-data access, background loading and unloading of bodies, collision queries running alongside simulation, and deterministic simulation by replicating inputs. The project also provides bindings for several languages.

JAX is a Python library for accelerator-oriented array computation and program transformation, designed for numerical computing and large-scale machine learning. It transforms Python and NumPy-style functions through automatic differentiation with `jax.grad`, automatic batching and vectorization with `jax.vmap`, and just-in-time compilation with `jax.jit`. It uses XLA to compile computations for GPUs, TPUs, and other accelerators. JAX supports compiler-driven multi-device execution: `Mesh`, `PartitionSpec`, and `NamedSharding` declare array placement across device meshes, allowing distributed training code to include automatically managed parallel work such as gradient synchronization. Its compilation tools also include ahead-of-time compilation, portable artifact export with `jax.export`, and conversion to TensorFlow SavedModels through `jax2tf` for serving integrations. The project describes itself as a research project rather than an official Google product.

An open-source set of reinforcement-learning training environments for Pollen Robotics' Microduck bipedal robot, built on mjlab with MuJoCo Warp and PPO. It trains walking, standing, recovery, sitting, ground-picking, kicking, rolling, and roller-skating policies, with flat, rough, slope, and backlash variants where applicable. The environments encode a sim-to-real pipeline using BAM actuator physics for Dynamixel XL330 servos, domain randomization for battery voltage, voltage sag, command delay, and friction, and simulated gear backlash with output-side encoder feedback. Policies use a shared 61-dimensional observation contract for runtime hot-swapping, are trained at 50 Hz, exported to ONNX with the observation normalizer embedded, and deployed by the Microduck runtime. Training runs locally through a CUDA GPU or can be submitted to Hugging Face Jobs; the project is licensed under Apache 2.0, while its 3D model files use CC BY-SA-NC.

jcode is an open-source, Rust-based terminal harness for AI coding agents and LLM workflows, designed for interactive development across multiple sessions. It supports resumable sessions, session search, multi-agent swarm and helper-agent workflows, compatible model-provider integrations, MCP, browser automation, and a self-development mode; installation scripts target macOS, Linux, and Windows. For automatic memory, jcode embeds turns and responses as semantic vectors, searches a graph of stored memories using cosine similarity, and feeds relevant results into the conversation. A memory sideagent can verify and expand retrieval, while periodic extraction stores new memories and ambient consolidation reorganizes entries and checks for staleness and conflicts. Explicit memory tools and traditional retrieval over previous sessions are also available. Its terminal UI includes side panels, diff views, inline Mermaid diagrams, information widgets, custom scrolling, and real-time rendering. The project supplies a Mermaid renderer without browser or TypeScript dependencies and publishes benchmarks covering RAM consumption, startup and input-readiness times, and memory scaling across multiple active sessions.

MiniMind is an open-source tiny large-language-model project and end-to-end training tutorial developed by Jingyao Gong. Its main dense model is approximately 64M parameters, and the repository provides the model architecture, tokenizer, datasets, inference code, and training pipeline. The pipeline covers pretraining, supervised fine-tuning, hand-written LoRA, DPO, PPO, GRPO, CISPO, model distillation, tool calling, adaptive thinking, and agentic reinforcement learning. Core algorithms are implemented directly with native PyTorch rather than relying on high-level abstractions from third-party training libraries. Its agentic reinforcement-learning path performs multi-turn rollouts, executes generated tool calls, appends tool observations to the context, calculates trajectory-level rewards, and updates the policy; rollout can use local PyTorch generation or an SGLang server. MiniMind supports dense and mixture-of-experts variants, single- and multi-GPU training, YaRN-based RoPE length extrapolation, Transformers-format models, and inference through llama.cpp, vLLM, or Ollama. The repository also includes a Streamlit chat interface and a lightweight OpenAI-compatible API server with tool-call and reasoning fields. It is released under the Apache License 2.0.