Tag
Explore by tags
Train robot reinforcement-learning agents with a heterogeneous runtime that streams CPU-parallel physics simulations (MuJoCo / Motrix) via shared memory into GPU/accelerator policy learners; provides a unified CLI, cross-platform backend support and demo checkpoints.
Provides 6,000 runnable, operator-level PyTorch tasks for training and evaluating CUDA kernel generation models; each sample includes executable code, operator descriptors, and provenance tags, with execution-driven filtering to ensure reproducibility and contamination control.
Large-scale in-the-wild robot manipulation dataset with ~76K teleoperated trajectories (~350 hours) that provides synchronized multi-view video, depth, camera calibration, robot state/action traces, and natural-language task instructions to train and evaluate manipulation policies and dynamics models. Collected across 564 scenes, 86 tasks, 52 buildings, on a uniform Franka Panda hardware stack and released in LeRobotDataset v3.0 format (≈707 GB, OpenMDW1.1).
Instruction-tuned 13B LLM post-trained on 260B tokens of pre-1931 English and finetuned with online DPO (LLM-as-judge) to improve instruction-following; suited for period-style English generation and etiquette/letter-writing formats, but not optimized for contemporary factual updates.
Provides 1.7M agent interaction traces in terminus-2 format for training and evaluating agentic LLMs and RL agents. Compiled from 219 source datasets across code repair, shell, math, competitive programming and general tasks; produced with the Harbor harness.
Aggregates 750k+ Harbor-compatible agentic tasks from 100+ public sources (Parquet shards preserved). Includes tasks with and without verifiers for RL evaluation or SFT/datagen workflows, enabling reproducible trace generation.
Preview of an MoE model family (V4-Pro: 1.6T params, 49B active; V4-Flash: 284B, 13B active) built for 1M-token contexts. A hybrid attention design cuts single-token inference FLOPs to 27% and KV cache to 10% versus V3.2 at million-token length.
Runs coding and long-running research workflows inside a persistent IPython environment with programmatic subagents and a durable 'Continual Harness' for session-level refinements. Key features include recursive subagents (RLM), executable Python skills, background daemon sessions, and evidence-backed local refinements. Best for reproducible, long-horizon coding, experiments, and evaluation pipelines where auditable agent-driven updates matter.
Provides tick-aligned Counter-Strike 2 player POV video clips with per-tick inputs and world-state sidecars — near-lossless 1280×720@32fps video, per-player stereo audio, and parquet indexes for event/kill/round filtering; suited for RL, video classification and clip mining.
Multimodal STEM problem set for verifiable, answer-supervised training and RL: contains single-image, multi-panel, and multi-image PhD-level questions across physics, math, chemistry and biology. Each example has a deterministic ground-truth answer, enabling reward modeling and automated evaluation.
Pairs natural-language instructions with executable setup artifacts and Python reward functions to create verifiable computer-use agent tasks. Provides a Parquet task table for fast filtering plus a compressed archive of runnable task bundles; several web task endpoints are placeholders that require a local CUA-Gym-Hub deployment.
Research-focused text-to-image foundation model that prioritizes training efficiency: a 3.8B-parameter architecture trained on an 800M image-text corpus with mixed-resolution learning, FLUX.2 VAE, RL tuning, and a distilled 4-step Lens-Turbo for fast high-resolution generation.