[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-84105":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":9,"totalLinesOfCode":9,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":9,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":21,"hasPages":21,"topics":23,"createdAt":9,"pushedAt":9,"updatedAt":29,"readmeContent":30,"aiSummary":9,"trendingCount":16,"starSnapshotCount":16,"syncStatus":31,"lastSyncTime":32,"discoverSource":33},84105,"magenta-realtime","magenta\u002Fmagenta-realtime","magenta","Magenta RealTime 2: An Open-Weights Live Music Model",null,"https:\u002F\u002Fgithub.com\u002Fmagenta\u002Fmagenta-realtime","Python",1512,163,21,10,0,46,72,164,98.84,false,"main",[24,25,26,27,28],"jax","machine-learning","mlx","music","real-time","2026-06-12 04:01:42","# Magenta RealTime 2\n\n[![CLI Tests](https:\u002F\u002Fgithub.com\u002Fmagenta\u002Fmagenta-realtime\u002Factions\u002Fworkflows\u002Fcli_tests.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmagenta\u002Fmagenta-realtime\u002Factions\u002Fworkflows\u002Fcli_tests.yml)\n\n> [!NOTE]\n> **Go [here](https:\u002F\u002Fmagenta.withgoogle.com\u002Fmrt2) for pre-built Apps & Plugins.**\n\nMagenta RealTime 2 (MRT2) is a state-of-the-art open-weights model for real-time music generation. It contains several key components\n* An [open-weights model](https:\u002F\u002Fhuggingface.co\u002Fgoogle\u002Fmagenta-realtime-2)\n* A [Python library](README.md) `magenta-rt` for inference with JAX and MLX backends\n* A [C++ inference engine](core\u002FREADME.md) `magentart::core` for efficient streaming audio generation on Apple Silicon MacBooks\n* A suite of [example applications](examples\u002FREADME.md) built on the inference engine.\n\nUse this project to run inference, build a DAW plugin, or embed the model into new applications of your imagination.\nFuture updates will support supervised fine-tuning.\n\n📖 **Full documentation:** https:\u002F\u002Fmagenta.github.io\u002Fmagenta-realtime\u002F\n(or build it locally — see [`docs\u002FREADME.md`](docs\u002FREADME.md)).\n\n> [!NOTE]\n> **Looking for Magenta RealTime v1?** The original model and code have been moved to the [`v1_legacy`](https:\u002F\u002Fgithub.com\u002Fmagenta\u002Fmagenta-realtime\u002Ftree\u002Fv1_legacy) branch.\n\n## Repo Highlights\n\n- `magenta_rt\u002F` — Python inference library (JAX \u002F MLX backends).\n- `core\u002F` — C++ inference library (`magentart::core`).\n- `examples\u002Fmrt2\u002Fauv3` — All-in-one AUv3 plugin for DAWs.\n- `examples\u002Fmrt2\u002Fstandalone` — All-in-one standalone macOS app.\n- `examples\u002Fjam\u002F` — App for exploring note control.\n- `examples\u002Fcollider\u002F` — App for exploring prompt space.\n- `notebooks\u002F` - Notebook for trying Python API.\n\n## Hardware requirements\n\n**Real-time streaming** requires **Apple Silicon** (M-series). We offer two model sizes:\n\n- **`mrt2_small`** (230M parameters) — runs real-time on any Apple Silicon Mac, including Air models.\n- **`mrt2_base`** (2.4B parameters) — higher quality; requires a Pro Max chip for real-time streaming.\n\nThe table below shows which devices support **real-time streaming** (generating audio faster than playback):\n\n| Device | `mrt2_small` (230M) | `mrt2_base` (2.4B) |\n|---|---|---|\n| M5 Max | ✅ | ✅ |\n| M3 Max | ✅ | ✅ |\n| M2 Max | ✅ | ✅ |\n| M4 Pro | ✅ | ✅ |\n| M2 Pro | ✅ | ❌ |\n| M1 Pro | ✅ | ❌ |\n| M4 Air | ✅ | ❌ |\n| M3 Air | ✅ | ❌ |\n| M1 Air | ✅ | ❌ |\n\n> **Note:** Both models can also run **offline (non-real-time) inference** on any Apple Silicon Mac or NVIDIA GPU via the Python library. See more details on [`docs\u002Fmodels.md`](docs\u002Fmodels.md).\n\n## Quickstart on Apple Silicon\n\n```bash\n# Install uv if you haven't and create a venv\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\nuv venv --python 3.12\nsource .venv\u002Fbin\u002Factivate\n\n# Install dependencies (Python dev)\nuv pip install \"magenta-rt[mlx]\"\n\n# Download resources: style model and codec model\n# (i.e., MusicCoCa and SpectroStream)\nmrt models init\n# Download the streaming model you want to use\nmrt models download\n# Generate 4 seconds of music (change to `mrt2_small` for small model)\nmrt mlx generate --prompt \"disco funk\" --duration 4.0 --model=mrt2_base\n```\n\n### Python Development\n\nFor local development, clone the repo instead of installing from PyPI:\n\n```bash\ngit clone --recurse-submodules https:\u002F\u002Fgithub.com\u002Fmagenta\u002Fmagenta-realtime.git\ncd magenta-realtime\nuv pip install -e \".[mlx]\"\n```\n\n### C++ App Development\n\nTo get started building C++ apps, perform the following setup:\n\n```bash\n# Install dependencies\nuv pip install \"cmake\u003C3.28\"\n\n# Build hello_mrt2 (a basic command-line interface)\ncmake . -B build\ncmake --build build --target hello_mrt2 -j10\n\n# Generate 4 seconds of music\n.\u002Fbuild\u002Fexamples\u002Fhello_mrt2\u002Fhello_mrt2 \\\n    ~\u002FDocuments\u002FMagenta\u002Fmagenta-rt-v2\u002Fmodels\u002Fmrt2_base\u002Fmrt2_base.mlxfn \\\n    ~\u002FDocuments\u002FMagenta\u002Fmagenta-rt-v2\u002Fresources \\\n    100 \\\n    --prompt \"ambient pads with sub bass\"\n```\n\nSee the full documentation:\n\n- [Installation](docs\u002Finstallation.md)\n- [Models & checkpoints](docs\u002Fmodels.md)\n- [Inference](docs\u002Finference.md)\n- [Exporting models](docs\u002Fexporting.md)\n- [Latency benchmark](docs\u002Fbenchmark.md)\n- [Testing](docs\u002Ftesting.md)\n\n## Other resources\n\n- [Get Started](https:\u002F\u002Fmagenta.withgoogle.com\u002Fmrt2)\n- [Blog Post](https:\u002F\u002Fmagenta.withgoogle.com\u002Fmagenta-realtime-2)\n- [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fgoogle\u002Fmagenta-realtime-2)\n\n## Citing this work\n\nPlease cite our previous [technical report](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.04651):\n\n**BibTeX:**\n\n```\n@article{gdmlyria2025live,\n    title={Live Music Models},\n    author={Caillon, Antoine and McWilliams, Brian and Tarakajian, Cassie and Simon, Ian and Manco, Ilaria and Engel, Jesse and Constant, Noah and Li, Pen and Denk, Timo I. and Lalama, Alberto and Agostinelli, Andrea and Huang, Anna and Manilow, Ethan and Brower, George and Erdogan, Hakan and Lei, Heidi and Rolnick, Itai and Grishchenko, Ivan and Orsini, Manu and Kastelic, Matej and Zuluaga, Mauricio and Verzetti, Mauro and Dooley, Michael and Skopek, Ondrej and Ferrer, Rafael and Borsos, Zal{\\'a}n and van den Oord, {\\\"A}aron and Eck, Douglas and Collins, Eli and Baldridge, Jason and Hume, Tom and Donahue, Chris and Han, Kehang and Roberts, Adam},\n    journal={arXiv:2508.04651},\n    year={2025}\n}\n```\n",2,"2026-06-11 04:12:18","trending"]