|
Download README.md from bageldotcom/Bagel-Robotics-1: direct link, hf CLI and curl.
- Browser
- Download file 7.34 kB
-
https://huggingface.co/bageldotcom/Bagel-Robotics-1/resolve/main/README.md
- Command line
-
hf download hf://bageldotcom/Bagel-Robotics-1/README.md
-
curl -L -o README.md https://huggingface.co/bageldotcom/Bagel-Robotics-1/resolve/main/README.md
7.34 kB
| language: en | |
| tags: | |
| - robotics | |
| - structured-pruning | |
| - robodojo | |
| - '#physicalai' | |
| - '#wam' | |
| # Bagel-Robotics-1 | |
| An open-weight world-action model for robot manipulation from Bagel Labs. | |
| <p align="center"> | |
| <a href="https://github.com/bageldotcom/Bagel-Robotics-1"><img alt="GitHub reference implementation" src="https://img.shields.io/badge/GitHub-Reference_Implementation-202936"></a> | |
| <a href="https://robodojo-benchmark.com/leaderboard?bench=sim"><img alt="RoboDojo-Sim leaderboard" src="https://img.shields.io/badge/RoboDojo--Sim-Leaderboard-365F98"></a> | |
| <a href="https://www.bagel.com/#real-world-demo"><img alt="Bagel Labs robot demos" src="https://img.shields.io/badge/Bagel_Labs-Robot_Demos-3B6055"></a> | |
| </p> | |
| | π₯ RoboDojo-Sim | βοΈ Compact video expert | π¦ Public release | | |
| | :---: | :---: | :---: | | |
| | **3rd overall** among standalone models by partial score | **12.468B parameters**, down **23.95%** | Checkpoint weights and inference code | | |
| *RoboDojo-Sim ranking as of November 20, 2026.* | |
| [Download the checkpoint](#download-the-exact-checkpoint) Β· [Run inference](#inference-adapter-and-reevaluation) Β· [Read the method](https://github.com/bageldotcom/Bagel-Robotics-1/blob/main/docs/METHOD.md) | |
| ## π¬ Robot demos | |
| | Coffee preparation | Tube placement | | |
| | :---: | :---: | | |
| | [](https://www.bagel.com/#real-world-demo) | [](https://www.bagel.com/#tube-placement-demo) | | |
| [Watch both demos on the Bagel Labs website](https://www.bagel.com/#real-world-demo). | |
| ## β¨ Overview | |
| Bagel-Robotics-1 uses **Data-Efficient Adaptive Pruning (DEAP)** to select informative windows from robot demonstrations, prune the video expert, and recover action quality by training the smaller model. | |
|  | |
| The video expert has **12,468,409,664 parameters**, 23.95% fewer than its pretrained starting point. Learned masks were converted into smaller FFN and cross-attention layers. Self-attention heads and the action expert are retained. Inference uses the exported model directly and requires neither teacher weights nor a learned-mask module. | |
| ## π Official RoboDojo-Sim results | |
| Official [RoboDojo-Sim leaderboard](https://robodojo-benchmark.com/leaderboard?bench=sim) results, **as of November 20, 2026**, excluding the Agent track: | |
| | Dimension | Partial score /100 | Success rate (%) | Ranking highlight | | |
| | --- | ---: | ---: | --- | | |
| | **Overall average** | **38.65** | **31.91** | π₯ **3rd** by partial score | | |
| | Average without Open | 47.33 | 39.05 | β | | |
| | Generalization | 35.52 | 28.56 | β | | |
| | **Generalization Standard (Gen-Std)** | **52.45** | **46.44** | π₯ **2nd** by partial score Β· π₯ **Tied 1st** by success rate | | |
| | Generalization Random (Gen-Rand) | 18.59 | 10.67 | β | | |
| | Precision | 42.65 | 35.42 | β | | |
| | Long-Horizon | 53.94 | 35.33 | β | | |
| | **Memory** | **57.21** | **56.89** | π₯ **2nd** by partial score | | |
| | Open | 3.93 | 3.33 | β | | |
| The overall average weights Generalization, Precision, Long-Horizon, Memory and Open equally. The average without Open uses the first four dimensions. Rankings are a dated leaderboard snapshot; refer to the official source for current standings. | |
| See the [results summary](https://github.com/bageldotcom/Bagel-Robotics-1/blob/main/results/summary.json) for machine-readable scores and rankings, and the [evaluation guide](https://github.com/bageldotcom/Bagel-Robotics-1/blob/main/docs/EVALUATION.md#reported-results) for settings and comparison requirements. | |
| ## Fixed checkpoint identity | |
| - Original checkpoint revision: `2651631c5d08819088475e8778676f5cf6d9afcf`. | |
| - Identity: `790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c`. | |
| - [`artifact-manifest.json`](artifact-manifest.json) pins all six bundle files and the separate public encoder/tokenizer assets. | |
| - [`SHA256SUMS`](SHA256SUMS) records the unchanged SHA-256 hashes of all six bundle files. | |
| The release adds documentation and inference packaging without changing any checkpoint file. The original revision is preserved, so historical provenance remains accessible in Git history. `COMPLETE.json` is the original export receipt, retained unchanged for integrity and not used as a current evaluation-status claim. | |
| ## Download the exact checkpoint | |
| The checkpoint and inference files are public and ungated. You can download them without a Hugging Face account or access token. | |
| ```bash | |
| hf download bageldotcom/Bagel-Robotics-1 \ | |
| --revision 2651631c5d08819088475e8778676f5cf6d9afcf \ | |
| --include video.pt action.pt compact_architecture.json dataset_stats.json COMPLETE.json validation.json \ | |
| --local-dir /absolute/path/to/checkpoints/cp1000 | |
| ``` | |
| There are approximately 29.0 GB of checkpoint tensors. Keep all six files together. The inference downloader checks their sizes and SHA-256 hashes and downloads approximately 16.7 GB of separately pinned VAE/conditioning/tokenizer assets. | |
| ## Inference adapter and reevaluation | |
| The standalone XPolicyLab adapter is provided under [`inference/`](inference/). The current public adapter follows the instructions in [`inference/README.md`](inference/README.md); no private research repository or access token is required. | |
| ```bash | |
| hf download bageldotcom/Bagel-Robotics-1 \ | |
| --include 'inference/**' \ | |
| --local-dir /absolute/path/to/Bagel-Robotics-1-release | |
| # Then install the adapter into an existing XPolicyLab checkout. | |
| python /absolute/path/to/Bagel-Robotics-1-release/inference/install_adapter.py /absolute/path/to/workspace/XPolicyLab | |
| ``` | |
| Follow [`inference/policy/Bagel_Robotics_1/README.md`](inference/policy/Bagel_Robotics_1/README.md) to install the Python 3.10 environment, download and verify assets, and run both debug checks. Follow [`REEVALUATION.md`](inference/policy/Bagel_Robotics_1/REEVALUATION.md) for task, suite and three-seed commands. Install RoboDojo and simulator assets separately using its [official guide](https://robodojo-benchmark.com/doc/usage/install-and-download/). | |
| The adapter uses `policy_name=Bagel_Robotics_1`, `ckpt_name=cp1000`, ARX X5 end-effector actions and one environment per policy process. Fixed inference settings: BF16; predict 32 actions and execute 16; ten Euler steps; shift 1; 90% new / 10% previous temporal blending. | |
| ## Validation and release scope | |
| The adapter qualification includes 15 CPU tests, both real-checkpoint debug transports, four valid native simulator episodes and eight exact action/history/reset/backend checks on tested recordings. | |
| The training, pruning and export implementation is publicly available in the [Bagel-Robotics-1 GitHub repository](https://github.com/bageldotcom/Bagel-Robotics-1). The technical report and citation are forthcoming. This Hugging Face repository contains the inference release and weights; it does not claim that optimizer state or exact training continuation can be reconstructed from weights alone. | |
| ## Attribution | |
| See the [source and artifact notices](THIRD_PARTY_NOTICES.md) for upstream credits, source provenance and license terms. Thanks to the RoboDojo and XPolicyLab teams for the benchmark and evaluation framework. | |