[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-92230":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":19,"topics":21,"createdAt":10,"pushedAt":10,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":15,"starSnapshotCount":15,"syncStatus":25,"lastSyncTime":26,"discoverSource":27},92230,"mira","mira-wm\u002Fmira","mira-wm","Code for MIRA: Multiplayer Interactive World Models with Representation Autoencoders","https:\u002F\u002Fmira-wm.com\u002F",null,"Python",351,17,1,0,154,53.77,"Apache License 2.0",false,"main",[],"2026-07-22 04:02:05","## MIRA: Multiplayer Interactive World Models with Representation Autoencoders\n\n[![Live demo](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdemo-mira--wm.com-0091FF.svg)](https:\u002F\u002Fmira-wm.com) \n[![Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpaper-technical%20report-b31b1b.svg)](https:\u002F\u002Fgithub.com\u002Fmira-wm\u002Fmira) \n[![Dataset](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20dataset-rocket--science-yellow.svg)](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fkyutai\u002Frocket-science) \n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache%202.0-blue.svg)](LICENSE)\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fca82b521-3e53-4f1b-8b27-9126a49aa5ea\n\nMIRA is a real-time world model of Rocket League: a 5B parameters\nlatent diffusion model generates the video frame by frame from all four players' actions, so a full 2v2\nmatch can be played inside the model at 20 FPS on a single GPU. Play it live at\n[mira-wm.com](https:\u002F\u002Fmira-wm.com).\n\nThis is the official code release for the technical report:\u003Cbr>\n[MIRA: Multiplayer Interactive World Models with Representation Autoencoders](https:\u002F\u002Fgithub.com\u002Fmira-wm\u002Fmira), a collaboration between\u003Cbr>\n[General Intuition](https:\u002F\u002Fwww.generalintuition.com), [Kyutai](https:\u002F\u002Fkyutai.org), and [Epic Games](https:\u002F\u002Fwww.epicgames.com\u002Fsite\u002Fhome).\n\n\u003Cimg width=\"1491\" height=\"533\" alt=\"architecture\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F4f0fa7c7-aa27-48ed-9b08-4ebaade65f4d\" \u002F>\n\n```\n@article{mira2026,\n  title={{MIRA}: Multiplayer Interactive World Models with Representation Autoencoders},\n  author={Hu, Anthony and Volhejn, V{\\'a}clav and Ramanana Rahary, Adrien and Mulder, Chris and Makkar, Aditya and Royer, Am{\\'e}lie and Liao, Alyx and Orsini, Manu and Jelley, Adam and Alonso, Eloi and Laurent, Florian and Nor{\\'e}n, Fredrik and Swingos, James and H{\\\"u}nermann, Jan and Rollins, Kent and Hosseini, Lucas and Le Cauchois, Matthieu and Peter, Maxim and de Witte, Pim and Brown, Tim and Micheli, Vincent and B{\\\"o}hle, Moritz and de Marmiesse, Gabriel and Sharmanska, Viktoriia and Specia, Lucia and Black, Michael and P{\\'e}rez, Patrick},\n  year={2026}\n}\n```\n\n### Installation\n\n```bash\npixi run setup   # one-time: creates the environment and installs MIRA (requires an NVIDIA GPU)\npixi run test    # run the test suite\n```\n\nRequires [pixi](https:\u002F\u002Fpixi.sh) and torch >= 2.8 (installed for you).\n\n### Dataset\n\nEach sample is a 4-second window of a 2v2 match, captured from all four players' synchronised views.\nFor every frame of every view, the dataset provides the video frame, the player's keyboard action,\nand the game state (ball, cars, and score). The full dataset is on\nthe [HuggingFace Hub](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fkyutai\u002Frocket-science) (see the\n[dataset card](docs\u002Fdataset_card.md)).\n\n```python\nfrom mira.data import RocketScienceDataset\n\n# Download the test split. shards=1 pulls just one shard for a quick look; omit it for the full split.\nds = RocketScienceDataset.from_hub(\"kyutai\u002Frocket-science\", split=\"test\", shards=1)\n\n# Load the clips of the first match (16 frames each, sampled at 20 FPS) and take the first one.\nclips = ds.load_match(ds.match_ids()[0], clip_len=16, target_fps=20)\nclip = clips[0]\n\n# A clip holds P=4 synchronised player views over T=16 frames:\nclip.frames    # tensor (P, T, C, H, W) uint8 — the rendered video, one per player\nclip.actions   # tensor (P, T, 9)       int32 — multi-hot keyboard state, aligned to each frame\nclip.events    # list of game events overlapping this window (goals, boost pickups, ...)\nclip.physics   # per player, per frame: game state (ball, cars, score)\n```\n\nExplore it interactively (4-player grid, keyboard overlay, top-down overview of the game using the game state):\n\n```bash\npixi run explore\n```\n\n### Training\n\nEntry points under `scripts\u002F` are Hydra apps — override any config key as `key=value`. Multi-GPU runs\ngo through `torchrun`; single-GPU works too.\n\n```bash\n# Train the codec\npython scripts\u002Ftrain_codec.py dataset.train_index=\u002Fpath\u002Fto\u002Ftrain dataset.test_index=\u002Fpath\u002Fto\u002Ftest\n\n# Train the single-player world model\npython scripts\u002Ftrain_world_model.py \\\n    model.architecture.config.codec_checkpoint=\u002Fpath\u002Fto\u002Fcodec_ckpt \\\n    dataset.train_index=\u002Fpath\u002Fto\u002Ftrain dataset.test_index=\u002Fpath\u002Fto\u002Ftest\n\n# Train the 4-player world model, warm-started from a single-player checkpoint\npython scripts\u002Ftrain_world_model.py model=multi_wrapper_world_model dataset.n_players=4 \\\n    model.architecture.config.wm_config.codec_checkpoint=\u002Fpath\u002Fto\u002Fcodec_ckpt \\\n    run.finetune_from=\u002Fpath\u002Fto\u002Fsingle_player_ckpt \\\n    dataset.train_index=\u002Fpath\u002Fto\u002Ftrain dataset.test_index=\u002Fpath\u002Fto\u002Ftest\n```\n\nCodec training uses a frozen **DINOv3-L\u002F16** encoder whose weights are gated by Meta. Download\n`dinov3_vitl16_pretrain_lvd1689m-8aa4cbdd.pth` from the\n[DINOv3 page](https:\u002F\u002Fai.meta.com\u002Fresources\u002Fmodels-and-libraries\u002Fdinov3-downloads\u002F) and set\n`RS_DINO_WEIGHTS_DIR=\u002Fpath\u002Fto\u002Fweights`. World-model training and inference don't need it.\n\n### Evaluation\n\n```bash\npython scripts\u002Feval_world_model_offline.py \u002Fpath\u002Fto\u002Fcheckpoint-1000\u002Fcheckpoint.pth\n```\n\n### License\n\nApache License 2.0 — see [LICENSE](LICENSE).\n","MIRA 是一个面向 Rocket League 游戏的多人实时交互式世界模型，基于 50 亿参数的潜在扩散架构，支持四玩家同步动作输入并以 20 FPS 在单 GPU 上逐帧生成完整 2v2 比赛视频。其核心技术包括表示自编码器（RAE）与多视角动作-状态联合建模，实现端到端的游戏动态预测与渲染。项目专为游戏AI仿真、强化学习环境建模及交互式世界模型研究设计，适用于需要高保真、低延迟、多智能体协同建模的场景。",2,"2026-07-08 04:29:53","CREATED_QUERY"]