Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Multitask DiT Policy","local":"multitask-dit-policy","sections":[{"title":"Model Overview","local":"model-overview","sections":[],"depth":2},{"title":"Installation Requirements","local":"installation-requirements","sections":[],"depth":2},{"title":"Usage","local":"usage","sections":[],"depth":2},{"title":"Training","local":"training","sections":[{"title":"Basic Training Command","local":"basic-training-command","sections":[],"depth":3},{"title":"Recommended Hyperparameters and Dataset Details (30Hz Control Frequency)","local":"recommended-hyperparameters-and-dataset-details-30hz-control-frequency","sections":[],"depth":3},{"title":"Training Configuration Parameters","local":"training-configuration-parameters","sections":[{"title":"Objective Selection","local":"objective-selection","sections":[],"depth":4},{"title":"Transformer Architecture","local":"transformer-architecture","sections":[],"depth":4},{"title":"Vision Encoder Configuration","local":"vision-encoder-configuration","sections":[],"depth":4},{"title":"Text Encoder Configuration","local":"text-encoder-configuration","sections":[],"depth":4},{"title":"Learning Rate Configuration","local":"learning-rate-configuration","sections":[],"depth":4}],"depth":3},{"title":"Training Tuning Guidelines","local":"training-tuning-guidelines","sections":[{"title":"1. Flow Matching with Beta Sampling","local":"1-flow-matching-with-beta-sampling","sections":[],"depth":4},{"title":"2. Number of Transformer Layers","local":"2-number-of-transformer-layers","sections":[],"depth":4},{"title":"3. horizon Tuning","local":"3-horizon-tuning","sections":[],"depth":4},{"title":"4. n_action_steps Sensitivity","local":"4-nactionsteps-sensitivity","sections":[],"depth":4}],"depth":3},{"title":"Inference Tuning","local":"inference-tuning","sections":[],"depth":3},{"title":"Resuming Training","local":"resuming-training","sections":[],"depth":3}],"depth":2},{"title":"Common Failure Modes and Debugging","local":"common-failure-modes-and-debugging","sections":[{"title":"Idling / No Motion","local":"idling--no-motion","sections":[],"depth":3},{"title":"Executing the Wrong Task","local":"executing-the-wrong-task","sections":[],"depth":3},{"title":"Training Instability","local":"training-instability","sections":[],"depth":3}],"depth":2},{"title":"Performance Considerations","local":"performance-considerations","sections":[{"title":"GPU Requirements","local":"gpu-requirements","sections":[],"depth":3},{"title":"Batch Size Recommendations","local":"batch-size-recommendations","sections":[],"depth":3}],"depth":2},{"title":"Example: Training on Custom Dataset","local":"example-training-on-custom-dataset","sections":[],"depth":2},{"title":"Libero Results","local":"libero-results","sections":[],"depth":2},{"title":"References","local":"references","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/lerobot/pr_3313/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/entry/start.d3f1c0f3.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/scheduler.eb244325.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/singletons.1f33814c.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/index.3c23fb4b.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/paths.17f05d75.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/entry/app.04bb7687.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/preload-helper.b00aacbc.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/index.3fe63ad3.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/nodes/0.07fbe93e.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/nodes/37.caf055d6.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/CopyLLMTxtMenu.d0c64540.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.6453902c.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/CodeBlock.48dd2cc2.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Multitask DiT Policy","local":"multitask-dit-policy","sections":[{"title":"Model Overview","local":"model-overview","sections":[],"depth":2},{"title":"Installation Requirements","local":"installation-requirements","sections":[],"depth":2},{"title":"Usage","local":"usage","sections":[],"depth":2},{"title":"Training","local":"training","sections":[{"title":"Basic Training Command","local":"basic-training-command","sections":[],"depth":3},{"title":"Recommended Hyperparameters and Dataset Details (30Hz Control Frequency)","local":"recommended-hyperparameters-and-dataset-details-30hz-control-frequency","sections":[],"depth":3},{"title":"Training Configuration Parameters","local":"training-configuration-parameters","sections":[{"title":"Objective Selection","local":"objective-selection","sections":[],"depth":4},{"title":"Transformer Architecture","local":"transformer-architecture","sections":[],"depth":4},{"title":"Vision Encoder Configuration","local":"vision-encoder-configuration","sections":[],"depth":4},{"title":"Text Encoder Configuration","local":"text-encoder-configuration","sections":[],"depth":4},{"title":"Learning Rate Configuration","local":"learning-rate-configuration","sections":[],"depth":4}],"depth":3},{"title":"Training Tuning Guidelines","local":"training-tuning-guidelines","sections":[{"title":"1. Flow Matching with Beta Sampling","local":"1-flow-matching-with-beta-sampling","sections":[],"depth":4},{"title":"2. Number of Transformer Layers","local":"2-number-of-transformer-layers","sections":[],"depth":4},{"title":"3. horizon Tuning","local":"3-horizon-tuning","sections":[],"depth":4},{"title":"4. n_action_steps Sensitivity","local":"4-nactionsteps-sensitivity","sections":[],"depth":4}],"depth":3},{"title":"Inference Tuning","local":"inference-tuning","sections":[],"depth":3},{"title":"Resuming Training","local":"resuming-training","sections":[],"depth":3}],"depth":2},{"title":"Common Failure Modes and Debugging","local":"common-failure-modes-and-debugging","sections":[{"title":"Idling / No Motion","local":"idling--no-motion","sections":[],"depth":3},{"title":"Executing the Wrong Task","local":"executing-the-wrong-task","sections":[],"depth":3},{"title":"Training Instability","local":"training-instability","sections":[],"depth":3}],"depth":2},{"title":"Performance Considerations","local":"performance-considerations","sections":[{"title":"GPU Requirements","local":"gpu-requirements","sections":[],"depth":3},{"title":"Batch Size Recommendations","local":"batch-size-recommendations","sections":[],"depth":3}],"depth":2},{"title":"Example: Training on Custom Dataset","local":"example-training-on-custom-dataset","sections":[],"depth":2},{"title":"Libero Results","local":"libero-results","sections":[],"depth":2},{"title":"References","local":"references","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg></button></div> </div> <h1 class="relative group"><a id="multitask-dit-policy" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#multitask-dit-policy"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Multitask DiT Policy</span></h1> <p data-svelte-h="svelte-fq7wf0">Multitask Diffusion Transformer (DiT) Policy is an evolution of the original Diffusion Policy architecture, which leverages a large DiT with text and vision conditioning for multitask robot learning. This implementation supports both diffusion and flow matching objectives for action generation, enabling robots to perform diverse manipulation tasks conditioned on language instructions.</p> <h2 class="relative group"><a id="model-overview" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#model-overview"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Model Overview</span></h2> <p data-svelte-h="svelte-16tn8ko">The model uses:</p> <ul data-svelte-h="svelte-i31piz"><li><strong>CLIP Vision Encoder</strong>: Processes RGB images from multiple camera views</li> <li><strong>CLIP Text Encoder</strong>: Encodes language task instructions (frozen weights with learnable projection)</li> <li><strong>Diffusion Transformer</strong>: Predicts action sequences conditioned on observations and language</li> <li><strong>Two Objectives</strong>: Supports both diffusion (DDPM/DDIM) and flow matching for action generation</li></ul> <p data-svelte-h="svelte-1fx3ud6">This model is exciting because you can achieve extremely high dexterity, competitive with multi-billion parameter | |
| VLAs, with only ~450M parameters and significantly less training.</p> <h2 class="relative group"><a id="installation-requirements" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#installation-requirements"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Installation Requirements</span></h2> <p data-svelte-h="svelte-1f1uo6d">Multitask DiT Policy has additional dependencies. Install it with:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->pip install lerobot[multi_task_dit]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1a6ur5o">This will install all necessary dependencies including the HuggingFace Transformers library for CLIP models.</p> <h2 class="relative group"><a id="usage" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#usage"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Usage</span></h2> <p data-svelte-h="svelte-1upxn8q">To use Multitask DiT in your LeRobot configuration, specify the policy type as:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->policy.<span class="hljs-built_in">type</span>=multi_task_dit<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="training" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#training"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Training</span></h2> <h3 class="relative group"><a id="basic-training-command" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#basic-training-command"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Basic Training Command</span></h3> <p data-svelte-h="svelte-ier52d">Here’s a complete training command for training Multitask DiT on your dataset:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-train \ | |
| --dataset.repo_id=YOUR_DATASET \ | |
| --output_dir=./outputs/multitask_dit_training \ | |
| --batch_size=32 \ | |
| --steps=5000 \ | |
| --save_freq=500 \ | |
| --log_freq=100 \ | |
| --policy.type=multi_task_dit \ | |
| --policy.device=cuda \ | |
| --policy.repo_id=<span class="hljs-string">"HF_USER/multitask-dit-your-robot"</span> \ | |
| --wandb.enable=<span class="hljs-literal">true</span><!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="recommended-hyperparameters-and-dataset-details-30hz-control-frequency" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#recommended-hyperparameters-and-dataset-details-30hz-control-frequency"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Recommended Hyperparameters and Dataset Details (30Hz Control Frequency)</span></h3> <p data-svelte-h="svelte-1e57d3o">For reliable performance, start with these suggested default hyperparameters:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-train \ | |
| --dataset.repo_id=YOUR_DATASET \ | |
| --output_dir=./outputs/mutitask_dit_training \ | |
| --batch_size=320 \ | |
| --steps=30000 \ | |
| --policy.type=multi_task_dit \ | |
| --policy.device=cuda \ | |
| --policy.horizon=32 \ | |
| --policy.n_action_steps=24 \ | |
| --policy.objective=diffusion \ | |
| --policy.noise_scheduler_type=DDPM \ | |
| --policy.num_train_timesteps=100 \ | |
| --policy.repo_id=<span class="hljs-string">"HF_USER/multitask-dit-your-robot"</span> \ | |
| --wandb.enable=<span class="hljs-literal">true</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-11zdjd0"><strong>Key Parameters:</strong></p> <ul data-svelte-h="svelte-3mwnn"><li><strong>Batch Size</strong>: 192-320 - If you have access to a GPU that can support this, you will get the best training dynamics</li> <li><strong>Horizon</strong>: 32 - number of action steps to predict, ~1.0 sec at 30Hz</li> <li><strong>n_action_steps</strong>: 24 - ~0.8 seconds at 30Hz</li> <li><strong>Objective</strong>: <code>diffusion</code> - start with diffusion and experiment with flow matching if generation quality is poor</li> <li><strong>Training Steps</strong>: >30k steps recommended for a single task</li></ul> <h3 class="relative group"><a id="training-configuration-parameters" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#training-configuration-parameters"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Training Configuration Parameters</span></h3> <h4 class="relative group"><a id="objective-selection" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#objective-selection"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Objective Selection</span></h4> <p data-svelte-h="svelte-1ww4omm">Choose between diffusion and flow matching:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Diffusion objective (default)</span> | |
| --policy.objective=diffusion \ | |
| --policy.noise_scheduler_type=DDPM \ <span class="hljs-comment"># or "DDIM"</span> | |
| --policy.num_train_timesteps=100 \ | |
| --policy.num_inference_steps=10 \ <span class="hljs-comment"># For faster inference</span> | |
| --policy.beta_schedule=squaredcos_cap_v2 \ <span class="hljs-comment"># Noise schedule type</span> | |
| --policy.prediction_type=epsilon \ <span class="hljs-comment"># "epsilon" (predict noise) or "sample" (predict clean)</span> | |
| --policy.clip_sample=<span class="hljs-literal">true</span> \ <span class="hljs-comment"># Clip samples during denoising</span> | |
| --policy.clip_sample_range=1.0 <span class="hljs-comment"># Clipping range [-x, x]</span> | |
| <span class="hljs-comment"># Flow matching objective</span> | |
| --policy.objective=flow_matching \ | |
| --policy.timestep_sampling_strategy=beta \ <span class="hljs-comment"># or "uniform" | the beta sampling strategy performance appears much better in practice</span> | |
| --policy.num_integration_steps=100 \ | |
| --policy.integration_method=euler \ <span class="hljs-comment"># or "rk4"</span> | |
| --policy.sigma_min=0.0 <span class="hljs-comment"># Minimum noise in flow interpolation path</span><!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="transformer-architecture" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#transformer-architecture"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Transformer Architecture</span></h4> <p data-svelte-h="svelte-dvvbwx">Adjust model capacity based on dataset size:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Small datasets (< 100 examples)</span> | |
| --policy.num_layers=4 \ | |
| --policy.hidden_dim=512 \ | |
| --policy.num_heads=8 <span class="hljs-comment"># should ideally be hidden_dim // 64</span> | |
| <span class="hljs-comment"># Medium datasets (100-5k examples) - default</span> | |
| --policy.num_layers=6 \ | |
| --policy.hidden_dim=512 \ | |
| --policy.num_heads=8 <span class="hljs-comment"># should ideally be hidden_dim // 64</span> | |
| <span class="hljs-comment"># Large datasets (> 5k examples)</span> | |
| --policy.num_layers=8 \ | |
| --policy.hidden_dim=512 \ | |
| --policy.num_heads=8 <span class="hljs-comment"># should ideally be hidden_dim // 64</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-obo8zq"><strong>Positional Encoding Options:</strong></p> <p data-svelte-h="svelte-4c8o3s">The model supports two positional encoding methods for action sequences:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Rotary Position Embedding (RoPE) - default, recommended</span> | |
| --policy.use_rope=<span class="hljs-literal">true</span> \ | |
| --policy.rope_base=10000.0 <span class="hljs-comment"># Base frequency for RoPE</span> | |
| <span class="hljs-comment"># Absolute positional encoding</span> | |
| --policy.use_positional_encoding=<span class="hljs-literal">true</span> <span class="hljs-comment"># Disables RoPE when true</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-13czyw2"><strong>Other Transformer Parameters:</strong></p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->--policy.dropout=0.1 <span class="hljs-comment"># Dropout rate for DiT blocks (0.0-1.0)</span> | |
| --policy.timestep_embed_dim=256 <span class="hljs-comment"># Timestep embedding dimension</span><!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="vision-encoder-configuration" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#vision-encoder-configuration"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Vision Encoder Configuration</span></h4> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Use different CLIP model for more expressivity at the cost of inference time</span> | |
| <span class="hljs-comment"># experiment with larger or smaller models depending on the complexity of your tasks and size of dataset</span> | |
| --policy.vision_encoder_name=openai/clip-vit-large-patch14 | |
| <span class="hljs-comment"># Use separate vision encoder per camera</span> | |
| <span class="hljs-comment"># This may be useful when cameras have significantly different characteristics, but</span> | |
| <span class="hljs-comment"># be wary of increased VRAM footprint.</span> | |
| --policy.use_separate_rgb_encoder_per_camera=<span class="hljs-literal">true</span> | |
| <span class="hljs-comment"># Image preprocessing</span> | |
| --policy.image_resize_shape=[XXX,YYY] \ <span class="hljs-comment"># you may need to resize your images for inference speed ups</span> | |
| --policy.image_crop_shape=[224,224] \ | |
| --policy.image_crop_is_random=<span class="hljs-literal">true</span> <span class="hljs-comment"># Random during training, center at inference</span><!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="text-encoder-configuration" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#text-encoder-configuration"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Text Encoder Configuration</span></h4> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Use different CLIP text encoder model</span> | |
| <span class="hljs-comment"># same as vision: experiment with larger or smaller models depending on the</span> | |
| <span class="hljs-comment"># complexity of your tasks and size of dataset</span> | |
| --policy.text_encoder_name=openai/clip-vit-large-patch14<!-- HTML_TAG_END --></pre></div> <h4 class="relative group"><a id="learning-rate-configuration" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#learning-rate-configuration"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Learning Rate Configuration</span></h4> <p data-svelte-h="svelte-120t5y7">The vision encoder uses a separate learning rate multiplier, where 1/10th is suggested to be the ideal staritng point:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->--policy.optimizer_lr=2e-5 \ | |
| --policy.vision_encoder_lr_multiplier=0.1 <span class="hljs-comment"># Vision encoder LR = 0.1 * optimizer_lr</span><!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="training-tuning-guidelines" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#training-tuning-guidelines"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Training Tuning Guidelines</span></h3> <h4 class="relative group"><a id="1-flow-matching-with-beta-sampling" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#1-flow-matching-with-beta-sampling"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>1. Flow Matching with Beta Sampling</span></h4> <p data-svelte-h="svelte-1op8hfn">The original diffusion implementation here is based on the work described in <a href="https://arxiv.org/abs/2507.05331" rel="nofollow">TRI’s LBM paper</a></p> <p data-svelte-h="svelte-1nha6gt">Additionally, we have implemented a flow-matching objective, which is described at a high-level in <a href="https://bostondynamics.com/blog/large-behavior-models-atlas-find-new-footing/" rel="nofollow">Boston Dynamics blog post</a>.</p> <p data-svelte-h="svelte-crm1o9">Consider testing the flow-matching objective and evaluating performance differences for your task:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->--policy.objective=flow_matching \ | |
| --policy.timestep_sampling_strategy=beta \ | |
| --policy.timestep_sampling_alpha=1.5 \ | |
| --policy.timestep_sampling_beta=1.0 \ | |
| --policy.timestep_sampling_s=0.999<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-12bskm7">This hasn’t been shown to be a silver bullet across every user case, but it occasionally results in smoother and more consistent actions.</p> <h4 class="relative group"><a id="2-number-of-transformer-layers" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#2-number-of-transformer-layers"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>2. Number of Transformer Layers</span></h4> <p data-svelte-h="svelte-1an1rz">Match model capacity to your dataset size:</p> <ul data-svelte-h="svelte-pnkxp7"><li><strong>Small datasets</strong> (< 100 examples): Reduce to 4 layers</li> <li><strong>Large datasets</strong> (> 5k examples): Increase to 8 layers</li></ul> <h4 class="relative group"><a id="3-horizon-tuning" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#3-horizon-tuning"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>3. horizon Tuning</span></h4> <p data-svelte-h="svelte-j70a5a">The model can be sensitive to the horizon you choose. Start with around a 1 second horizon based on your control frequency:</p> <ul data-svelte-h="svelte-oglx66"><li><strong>30 Hz frequency</strong>: <code>horizon=30</code></li> <li><strong>10 Hz frequency</strong>: <code>horizon=10</code></li></ul> <p data-svelte-h="svelte-1r86l1r">Then experiment with increasing from there. The horizon determines how far into the future the model predicts actions.</p> <h4 class="relative group"><a id="4-nactionsteps-sensitivity" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#4-nactionsteps-sensitivity"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>4. n_action_steps Sensitivity</span></h4> <p data-svelte-h="svelte-1bfvpbi">The model can also be very sensitive to <code>n_action_steps</code>. Start with it being around 0.8 seconds based on your control frequency and tune from there:</p> <ul data-svelte-h="svelte-12z7a69"><li><strong>Lower values</strong>: More reactive but potentially less stable for long-horizon tasks</li> <li><strong>Higher values</strong>: Better for long-horizon execution but open-loop failures are limited in their recovery</li></ul> <h3 class="relative group"><a id="inference-tuning" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#inference-tuning"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Inference Tuning</span></h3> <p data-svelte-h="svelte-1ij28ip">For faster inference, use DDIM with fewer sampling steps:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->--policy.noise_scheduler_type=DDIM \ | |
| --policy.num_inference_steps=10<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="resuming-training" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#resuming-training"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Resuming Training</span></h3> <p data-svelte-h="svelte-hnhgb5">To resume training from a checkpoint:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-train \ | |
| --config_path=./outputs/mutitask_dit_training/checkpoints/last/pretrained_model/train_config.json \ | |
| --resume=<span class="hljs-literal">true</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1piqpt9">The checkpoint directory should contain <code>model.safetensors</code> and <code>config.json</code> files (saved automatically during training). When resuming, the configuration is loaded from the checkpoint, so you don’t need to specify other parameters.</p> <h2 class="relative group"><a id="common-failure-modes-and-debugging" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#common-failure-modes-and-debugging"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Common Failure Modes and Debugging</span></h2> <p data-svelte-h="svelte-cejt3t">Training these models can be finicky. Here are common failure modes and debugging approaches:</p> <h3 class="relative group"><a id="idling--no-motion" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#idling--no-motion"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Idling / No Motion</span></h3> <p data-svelte-h="svelte-jk6con">The model may “collapse” during inference, resulting in static or no motion. This can occur when:</p> <ol data-svelte-h="svelte-1j8npe0"><li><p><strong>Insufficient training data</strong>: If you only have 20-50 examples, try to roughly double your dataset size. Once you have above 300 examples, if you’re still seeing this, the task may be too complex.</p></li> <li><p><strong>Multiple similar tasks</strong>: When your dataset contains multiple similar tasks (e.g., picking up 2 different objects), the model may rely too heavily on language conditioning which might not be rich enough.</p></li></ol> <p data-svelte-h="svelte-1c76s7d"><strong>Debugging tips:</strong></p> <ul data-svelte-h="svelte-3vieto"><li>Increase dataset size (double until you get to over 300 examples)</li> <li>Train for longer, up to 100k steps, even when the loss flatlines</li> <li>Check if the model is receiving proper language instructions or increase diversity of instruction</li></ul> <h3 class="relative group"><a id="executing-the-wrong-task" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#executing-the-wrong-task"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Executing the Wrong Task</span></h3> <p data-svelte-h="svelte-15v2wnp">Sometimes the robot will completely ignore your instruction and perform some other task. This generally only happens if you have trained on multiple tasks.</p> <p data-svelte-h="svelte-16p9a9p"><strong>Potential causes:</strong></p> <ul data-svelte-h="svelte-o4w6gr"><li>Language instruction ambiguity</li> <li>Insufficient task-specific training data</li> <li>Model confusion between similar tasks in the multitask dataset</li></ul> <p data-svelte-h="svelte-1c76s7d"><strong>Debugging tips:</strong></p> <ul data-svelte-h="svelte-wdlbng"><li>Verify language instruction specificity, especially if descriptions are similar between multiple tasks</li> <li>Check task distribution in your training dataset and add weighting to the failing/ignored task</li> <li>Consider task-specific fine-tuning</li></ul> <h3 class="relative group"><a id="training-instability" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#training-instability"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Training Instability</span></h3> <p data-svelte-h="svelte-wbk49u">If training loss is unstable or diverging:</p> <ul data-svelte-h="svelte-1b02erw"><li>Try adjusting learning rate between <code>1e-5</code> and <code>3e-4</code></li> <li>Increase batch size if possible</li> <li>Check that your dataset normalization is correct</li> <li>Verify image preprocessing is working correctly</li></ul> <h2 class="relative group"><a id="performance-considerations" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#performance-considerations"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Performance Considerations</span></h2> <h3 class="relative group"><a id="gpu-requirements" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#gpu-requirements"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>GPU Requirements</span></h3> <ul data-svelte-h="svelte-s3kshv"><li><strong>Inference</strong>: At least an RTX 5070 Ti (or equivalent GPU) is recommended for reasonable speed performance</li> <li><strong>Training</strong>: A GPU with enough VRAM to load batch sizes of >64 is ideal, which will vary depending on the number of image observations, etc</li></ul> <h3 class="relative group"><a id="batch-size-recommendations" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#batch-size-recommendations"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Batch Size Recommendations</span></h3> <ul data-svelte-h="svelte-k281bs"><li><strong>Minimum</strong>: 64 (less than this may result in unstable training)</li> <li><strong>Recommended</strong>: 256-320 (best performance, requires larger GPU)</li></ul> <h2 class="relative group"><a id="example-training-on-custom-dataset" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#example-training-on-custom-dataset"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Example: Training on Custom Dataset</span></h2> <p data-svelte-h="svelte-1cp7go5">Here’s a complete example training on a custom dataset:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-train \ | |
| --dataset.repo_id=YOUR_DATASET \ | |
| --output_dir=./outputs/mutitask_dit_training \ | |
| --batch_size=320 \ | |
| --steps=30000 \ | |
| --save_freq=1000 \ | |
| --log_freq=100 \ | |
| --eval_freq=1000 \ | |
| --policy.type=multi_task_dit \ | |
| --policy.device=cuda \ | |
| --policy.horizon=32 \ | |
| --policy.n_action_steps=24 \ | |
| --policy.objective=diffusion \ | |
| --policy.noise_scheduler_type=DDPM \ | |
| --policy.num_layers=6 \ | |
| --policy.hidden_dim=512 \ | |
| --policy.vision_encoder_name=openai/clip-vit-base-patch16 \ | |
| --policy.image_resize_shape=[320,240] \ | |
| --policy.image_crop_shape=[224,224] \ | |
| --policy.repo_id=<span class="hljs-string">"HF_USER/multitask-dit-your-robot"</span> \ | |
| --wandb.enable=<span class="hljs-literal">true</span> \ | |
| --wandb.project=multitask_dit<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="libero-results" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#libero-results"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Libero Results</span></h2> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->python -m lerobot.scripts.lerobot_train \ | |
| <span class="hljs-params">--dataset</span>.repo_id=HuggingFaceVLA/libero \ | |
| <span class="hljs-params">--policy</span>.type=multi_task_dit \ | |
| <span class="hljs-params">--policy</span>.push_to_hub=<span class="hljs-literal">false</span> \ | |
| <span class="hljs-params">--output_dir=</span><span class="hljs-string">"./outputs/multitask_dit_libero"</span> \ | |
| <span class="hljs-params">--job_name=</span><span class="hljs-string">"multitask-dit-libero"</span> \ | |
| <span class="hljs-params">--wandb</span>.enable=<span class="hljs-literal">true</span> \ | |
| <span class="hljs-params">--wandb</span>.project=multitask_dit_libero \ | |
| <span class="hljs-params">--dataset</span>.image_transforms.enable=<span class="hljs-literal">true</span> \ | |
| <span class="hljs-params">--dataset</span>.image_transforms.max_num_transforms=4 \ | |
| <span class="hljs-params">--dataset</span>.image_transforms.tfs='{<span class="hljs-string">"brightness"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"ColorJitter"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"brightness"</span>:[0.75,1.25]}},<span class="hljs-string">"contrast"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"ColorJitter"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"contrast"</span>:[0.6,1.4]}},<span class="hljs-string">"saturation"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"ColorJitter"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"saturation"</span>:[0.8,1.2]}},<span class="hljs-string">"hue"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"ColorJitter"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"hue"</span>:[-0.05,0.05]}},<span class="hljs-string">"sharpness"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"SharpnessJitter"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"sharpness"</span>:[0.6,1.4]}},<span class="hljs-string">"rotation"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"RandomRotation"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"degrees"</span>:[-5,5]}},<span class="hljs-string">"translation"</span>:{<span class="hljs-string">"type"</span>:<span class="hljs-string">"RandomAffine"</span>,<span class="hljs-string">"kwargs"</span>:{<span class="hljs-string">"degrees"</span><span class="hljs-function">:0</span>,<span class="hljs-string">"translate"</span>:[0.1,0.1]}}}' \ | |
| <span class="hljs-params">--dataset</span>.video_backend=torchcodec \ | |
| <span class="hljs-params">--policy</span>.use_amp=<span class="hljs-literal">true</span> \ | |
| <span class="hljs-params">--policy</span>.horizon=48 \ | |
| <span class="hljs-params">--policy</span>.n_obs_steps=2 \ | |
| <span class="hljs-params">--policy</span>.use_rope=<span class="hljs-literal">true</span> \ | |
| <span class="hljs-params">--policy</span>.use_positional_encoding=<span class="hljs-literal">false</span> \ | |
| <span class="hljs-params">--policy</span>.hidden_dim=768 \ | |
| <span class="hljs-params">--policy</span>.num_layers=8 \ | |
| <span class="hljs-params">--policy</span>.num_heads=12 \ | |
| <span class="hljs-params">--policy</span>.dropout=0.1 \ | |
| <span class="hljs-params">--policy</span>.timestep_embed_dim=256 \ | |
| <span class="hljs-params">--policy</span>.objective=diffusion \ | |
| <span class="hljs-params">--policy</span>.optimizer_lr=3e-4 \ | |
| <span class="hljs-params">--policy</span>.optimizer_weight_decay=0 \ | |
| <span class="hljs-params">--policy</span>.scheduler_warmup_steps=0 \ | |
| <span class="hljs-params">--policy</span>.vision_encoder_name=openai/clip-vit-base-patch16 \ | |
| <span class="hljs-params">--policy</span>.image_resize_shape=[256,256] \ | |
| <span class="hljs-params">--policy</span>.image_crop_is_random=<span class="hljs-literal">true</span> \ | |
| <span class="hljs-params">--policy</span>.text_encoder_name=openai/clip-vit-base-patch16 \ | |
| <span class="hljs-params">--policy</span>.vision_encoder_lr_multiplier=0.1 \ | |
| <span class="hljs-params">--policy</span>.device=cuda \ | |
| <span class="hljs-params">--num_workers=8</span> \ | |
| <span class="hljs-params">--save_freq=4000</span> \ | |
| <span class="hljs-params">--log_freq=100</span> \ | |
| <span class="hljs-params">--steps=100000</span> \ | |
| <span class="hljs-params">--batch_size=320</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1kjucsw">Results:</p> <table data-svelte-h="svelte-1vrru3a"><thead><tr><th>LIBERO Spatial</th> <th>LIBERO Object</th> <th>LIBERO Goal</th> <th>LIBERO 10</th> <th>Average</th></tr></thead> <tbody><tr><td>87.0</td> <td>98.2</td> <td>93.8</td> <td>83.2</td> <td>90.6</td></tr></tbody></table> <h2 class="relative group"><a id="references" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#references"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>References</span></h2> <p data-svelte-h="svelte-x7z092">For more details on the technical implementation and architecture, see:</p> <ul data-svelte-h="svelte-roonp6"><li><a href="https://arxiv.org/abs/2507.05331" rel="nofollow">A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation</a></li> <li><a href="https://bostondynamics.com/blog/large-behavior-models-atlas-find-new-footing/" rel="nofollow">Large Behavior Models and Atlas Find New Footing</a></li> <li><a href="https://brysonkjones.substack.com/p/dissecting-and-open-sourcing-multitask-diffusion-transformer-policy" rel="nofollow">Dissecting and Open-Sourcing Multitask Diffusion Transformer Policy</a></li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/lerobot/blob/main/docs/source/multi_task_dit.mdx" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg> <span data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_9kza6s = { | |
| assets: "/docs/lerobot/pr_3313/en", | |
| base: "/docs/lerobot/pr_3313/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/lerobot/pr_3313/en/_app/immutable/entry/start.d3f1c0f3.js"), | |
| import("/docs/lerobot/pr_3313/en/_app/immutable/entry/app.04bb7687.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 37], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 87.1 kB
- Xet hash:
- e2d5c98cd393d62d5da07a101b5751b153097906f8f8bde6c97a645deaa40203
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.