trl/pr_5607/en/_app/immutable/nodes/35.9fed32cd.js

import{s as Ps,n as Ds,o as Os}from"../chunks/scheduler.7b731bd4.js";import{S as Ks,i as tl,e as i,s as l,c as u,h as el,a as r,d as e,b as a,f as g,g as m,j as p,k as zs,l as s,m as o,n as c,t as d,o as h,p as M}from"../chunks/index.cc268345.js";import{C as nl,H as T,E as sl}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.f0d99f98.js";import{C as y}from"../chunks/CodeBlock.169a125f.js";function ll(ns){let $,ve,Te,je,Y,be,x,we,V,ss="NVIDIA NeMo Gym is a library for building RL environments for large language models. This integration enables training models in NeMo Gym environments using TRL’s GRPOTrainer with vLLM server mode.",Ue,k,ls="The integration supports multi-step and multi-turn rollouts, multi-environment training, and any NeMo Gym environment (thoroughly tested: workplace assistant, reasoning gym, MCQA, and math with judge).",$e,H,_e,W,as="<li><strong>Production-Ready Scale</strong>: Tested for frontier model training with diverse environments running in parallel across math, coding, tool use, reasoning, and more.</li> <li><strong>Multi-Verifier Training</strong>: Supports algorithmic verification, LLM-as-a-judge, and custom verification logic in a single training run.</li> <li><strong>Decoupled Architecture</strong>: Build agents and environments independently from the training loop—no RL framework expertise required.</li> <li><strong>OpenAI-Compatible API</strong>: All environments use the standardized OpenAI Responses API for seamless integration with vLLM, OpenAI models, and other endpoints.</li>",Ce,S,Ie,A,is="NeMo Gym provides training-ready environments across multiple domains, including but not limited to:",Ze,F,rs="<thead><tr><th>Environment</th> <th>Domain</th> <th>Description</th></tr></thead> <tbody><tr><td>Workplace Assistant</td> <td>Agent</td> <td>Multi-step tool calling in common office scenarios (calendar, email, and more)</td></tr> <tr><td>Math with Judge</td> <td>Math</td> <td>Math problems with algorithmic or judge-based verification</td></tr> <tr><td>Code Gen</td> <td>Coding</td> <td>Competitive programming problems with code execution</td></tr> <tr><td>MCQA</td> <td>Knowledge</td> <td>Multiple-choice question answering</td></tr> <tr><td>Instruction Following</td> <td>Instruction Following</td> <td>IFEval/IFBench style tasks</td></tr> <tr><td>Reasoning Gym</td> <td>Multiple</td> <td>Single-step procedurally generated verifiable tasks across domains</td></tr></tbody>",Le,Q,os='For a complete list of available training environments, refer to the <a href="https://github.com/NVIDIA-NeMo/Gym#-available-resource-servers" rel="nofollow">NeMo Gym repository</a>.',Ge,q,Be,z,ps="Complete these one-time setup steps before running training.",Ne,P,Ee,G,D,Yt,us="<strong>Install TRL with vLLM extras</strong>",Jn,O,vn,K,xt,ms="<strong>Install NeMo Gym</strong>",jn,tt,Re,et,Xe,nt,cs="Many NeMo Gym datasets used to train Nemotron models are available on Hugging Face. Use <code>ng_prepare_data</code> to download and prepare datasets. This command:",Ye,st,ds="<li>Downloads the dataset from Hugging Face</li> <li>Validates the data format</li> <li>Adds an <code>agent_ref</code> field to each example that tells NeMo Gym which agent server should handle that example</li>",xe,lt,hs="<p><strong>Note</strong>: <code>train_multi_environment.py</code> adds the <code>agent_ref</code> field when loading datasets, so this step is optional if datasets are created another way.</p>",Ve,B,_,Vt,Ms="<strong>Set Hugging Face Token</strong>",bn,kt,gs="Create <code>env.yaml</code> in <code>Gym/</code> with your HF token:",wn,at,Un,C,Ht,ys="<strong>Prepare Dataset</strong>",$n,it,_n,Wt,fs="This creates <code>train.jsonl</code> and <code>validation.jsonl</code> files in <code>data/workplace_assistant/</code>.",ke,rt,Ts='To create a new environment, refer to the <a href="https://docs.nvidia.com/nemo/gym/latest/contribute/environments/new-environment.html" rel="nofollow">environment creation guide</a>. We suggest running an existing one first!',He,ot,We,pt,Js="NeMo Gym datasets are stored as JSONL. Each line contains a task with input messages, tool definitions, metadata such as ground truth for verification, and an agent server reference. The following example shows the workplace dataset structure. Metadata fields can differ between datasets, as long as the corresponding resources server uses the fields appropriately.",Se,ut,Ae,mt,Fe,ct,vs="For development and testing on a single node.",Qe,dt,qe,N,I,St,js="<strong>Update Environment Config</strong>",Cn,At,bs="Update <code>env.yaml</code> in <code>Gym/</code> to include model information:",In,ht,Zn,Ft,ws="<p><strong>Update Training Config</strong></p> <p>Update <code>examples/scripts/nemo_gym/config.yaml</code> to point to the dataset generated above, and any other optional modifications.</p>",ze,Mt,Pe,gt,Us="The following steps run in 3 terminals. It can also be ran with processes in the background, or using tmux.",De,b,J,Qt,$s="<strong>Start NeMo Gym Servers</strong> (Terminal 1)",Ln,yt,Gn,qt,_s="This starts:",Bn,zt,Cs="<li><strong>Agent server</strong>: Orchestrates rollouts using resource servers and model servers</li> <li><strong>Resources server</strong>: Supports environment logic such as state-management, tool implementations, and task verification</li> <li><strong>Model server</strong>: Adapts vLLM server requests to support NeMo Gym agents and on-policy RL training while ensuring OpenAI API compatibility</li> <li><strong>Head server</strong>: Manages servers used in training enabling their discovery</li>",Nn,ft,Pt,Is="<strong>Start TRL vLLM Server on GPU 0</strong> (Terminal 2)",En,Tt,Rn,Jt,Dt,Zs="<strong>Run Training on GPU 1</strong> (Terminal 3)",Xn,vt,Oe,jt,Ke,bt,Ls="An example five-node training script is provided in <code>submit.sh</code>. Nodes one through four run the training algorithm, while node five runs vLLM inference for NeMo Gym agent rollouts.",tn,w,Ot,Gs="<p><strong>Configure the Script</strong></p> <p>Update <code>submit.sh</code> with your Slurm account, partition, paths to your project directory, and updated training configs.</p>",Yn,wt,Kt,Bs="<strong>Submit the Job</strong>",xn,Ut,Vn,$t,te,Ns="<strong>Monitor Training</strong>",kn,_t,en,Ct,Es="<p><strong>Tip</strong>: Set up wandb logging for detailed training metrics. For more details on TRL’s vLLM integration, refer to the vLLM integration page.</p>",nn,It,sn,Zt,Rs="Train on multiple NeMo Gym environments simultaneously. This allows learning diverse capabilities, such as tool calling and math reasoning, in a single training run.",ln,f,Z,ee,Xs="<strong>Prepare Individual Datasets</strong>",Hn,ne,Ys="Prepare datasets for each environment. The workplace assistant dataset was prepared above. Now lets create a dataset for the mini sudoku environment implemented by the reasoning gym resources server in NeMo Gym:",Wn,Lt,Sn,v,se,xs="<strong>Create Combined Dataset</strong>",An,le,Vs="Combine datasets into a single file with tasks from both environments:",Fn,Gt,Qn,ae,ks="<p><strong>Tip</strong>: Ensure datasets are the same size before shuffling for an even blend of tasks. Repeat for the validation dataset.</p>",qn,L,ie,Hs="<strong>Update Training Config</strong>",zn,re,Ws="Update the config to point to the combined dataset:",Pn,Bt,Dn,j,oe,Ss="<strong>Update ng_run</strong>",On,pe,As="Whether training interactively or via Slurm, update the <code>ng_run</code> command to include config files from each resources server:",Kn,Nt,ts,ue,Fs="This starts servers for both environments. The training script automatically routes each example to the correct agent server based on its <code>agent_ref</code> field.",es,me,Qs="<p><strong>Run Training</strong></p> <p>Update the Slurm submission script to use the new training config and both <code>ng_run</code> resources server configs, then submit the job as before.</p> <p>The training script reads <code>agent_ref</code> from each example’s metadata, routes requests to the correct NeMo Gym agent server, and handles different agents and environments in the same batch.</p>",an,Et,rn,Rt,qs='<li><a href="https://github.com/NVIDIA-NeMo/Gym" rel="nofollow">NeMo Gym GitHub</a></li> <li><a href="https://docs.nvidia.com/nemo/gym/latest/" rel="nofollow">NeMo Gym Documentation</a></li> <li><a href="https://github.com/huggingface/trl/blob/main/examples/scripts/nemo_gym/train_multi_environment.py" rel="nofollow">Training Script</a></li> <li><a href="grpo_trainer">TRL GRPO Trainer</a></li>',on,Xt,pn,Je,un;return Y=new nl({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),x=new T({props:{title:"NeMo Gym Integration",local:"nemo-gym-integration",headingTag:"h1"}}),H=new T({props:{title:"Why NeMo Gym",local:"why-nemo-gym",headingTag:"h2"}}),S=new T({props:{title:"Available Environments",local:"available-environments",headingTag:"h2"}}),q=new T({props:{title:"Before You Start",local:"before-you-start",headingTag:"h2"}}),P=new T({props:{title:"Install TRL and NeMo Gym",local:"install-trl-and-nemo-gym",headingTag:"h3"}}),O=new y({props:{code:"Y2QlMjB0cmwlMkYlMEF1diUyMHZlbnYlMEFzb3VyY2UlMjAudmVudiUyRmJpbiUyRmFjdGl2YXRlJTBBdXYlMjBzeW5jJTIwLS1leHRyYSUyMHZsbG0=",highlighted:`<span class="hljs-built_in">cd</span> trl/
uv venv
<span class="hljs-built_in">source</span> .venv/bin/activate
uv <span class="hljs-built_in">sync</span> --extra vllm`,wrap:!1}}),tt=new y({props:{code:"JTIzJTIwZGVhY3RpdmF0ZSUyMHRybCUyMHZlbnYlMEFkZWFjdGl2YXRlJTBBZ2l0JTIwY2xvbmUlMjBodHRwcyUzQSUyRiUyRmdpdGh1Yi5jb20lMkZOVklESUEtTmVNbyUyRkd5bS5naXQlMEFjZCUyMEd5bSUwQXV2JTIwdmVudiUyMC0tcHl0aG9uJTIwMy4xMiUwQXNvdXJjZSUyMC52ZW52JTJGYmluJTJGYWN0aXZhdGUlMEF1diUyMHN5bmM=",highlighted:`<span class="hljs-comment"># deactivate trl venv</span>
deactivate
git <span class="hljs-built_in">clone</span> https://github.com/NVIDIA-NeMo/Gym.git
<span class="hljs-built_in">cd</span> Gym
uv venv --python 3.12
<span class="hljs-built_in">source</span> .venv/bin/activate
uv <span class="hljs-built_in">sync</span>`,wrap:!1}}),et=new T({props:{title:"Prepare a Dataset",local:"prepare-a-dataset",headingTag:"h3"}}),at=new y({props:{code:"aGZfdG9rZW4lM0ElMjAlM0N5b3VyX2hmX3Rva2VuJTNF",highlighted:'<span class="hljs-attr">hf_token:</span> <span class="hljs-string">&lt;your_hf_token&gt;</span>',wrap:!1}}),it=new y({props:{code:"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",highlighted:`<span class="hljs-comment"># Enter Gym and activate the venv</span>
<span class="hljs-built_in">cd</span> Gym
<span class="hljs-built_in">source</span> .venv/bin/activate

<span class="hljs-comment"># Set config paths</span>
config_paths=<span class="hljs-string">&quot;responses_api_models/vllm_model/configs/vllm_model.yaml,\\
resources_servers/workplace_assistant/configs/workplace_assistant.yaml&quot;</span>

<span class="hljs-comment"># Download data and prep for training</span>
ng_prepare_data <span class="hljs-string">&quot;+config_paths=[<span class="hljs-variable">\${config_paths}</span>]&quot;</span> \\
    +output_dirpath=data/workplace_assistant \\
    +mode=train_preparation \\
    +should_download=<span class="hljs-literal">true</span> \\
    +data_source=huggingface`,wrap:!1}}),ot=new T({props:{title:"Dataset Format",local:"dataset-format",headingTag:"h4"}}),ut=new y({props:{code:"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",highlighted:`<span class="hljs-punctuation">{</span>
  <span class="hljs-attr">&quot;responses_create_params&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
    <span class="hljs-attr">&quot;input&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span>
      <span class="hljs-punctuation">{</span><span class="hljs-attr">&quot;role&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;system&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;content&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;...&quot;</span><span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
      <span class="hljs-punctuation">{</span><span class="hljs-attr">&quot;role&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;user&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;content&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;Move any of jinsoo&#x27;s tasks that are in review to completed&quot;</span><span class="hljs-punctuation">}</span>
    <span class="hljs-punctuation">]</span><span class="hljs-punctuation">,</span>
    <span class="hljs-attr">&quot;tools&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span>...<span class="hljs-punctuation">]</span><span class="hljs-punctuation">,</span>
    <span class="hljs-attr">&quot;parallel_tool_calls&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">false</span></span><span class="hljs-punctuation">,</span>
    <span class="hljs-attr">&quot;temperature&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">1</span>
  <span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
  <span class="hljs-attr">&quot;ground_truth&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span>
    <span class="hljs-punctuation">{</span><span class="hljs-attr">&quot;name&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;project_management_update_task&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;arguments&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;{...}&quot;</span><span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
    ...
  <span class="hljs-punctuation">]</span><span class="hljs-punctuation">,</span>
  <span class="hljs-attr">&quot;category&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;workbench_project_management&quot;</span><span class="hljs-punctuation">,</span>
  <span class="hljs-attr">&quot;environment_name&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;workbench&quot;</span><span class="hljs-punctuation">,</span>
  <span class="hljs-attr">&quot;agent_ref&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
    <span class="hljs-attr">&quot;type&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;responses_api_agents&quot;</span><span class="hljs-punctuation">,</span>
    <span class="hljs-attr">&quot;name&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;workplace_assistant_simple_agent&quot;</span>
  <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span>`,wrap:!1}}),mt=new T({props:{title:"Interactive Training",local:"interactive-training",headingTag:"h2"}}),dt=new T({props:{title:"Set Up",local:"set-up",headingTag:"h3"}}),ht=new y({props:{code:"cG9saWN5X2Jhc2VfdXJsJTNBJTIwaHR0cCUzQSUyRiUyRjEyNy4wLjAuMSUzQTgwMDAlMkZ2MSUwQXBvbGljeV9hcGlfa2V5JTNBJTIwRU1QVFklMEFwb2xpY3lfbW9kZWxfbmFtZSUzQSUyMFF3ZW4lMkZRd2VuMi41LTEuNUItSW5zdHJ1Y3QlMEFoZl90b2tlbiUzQSUyMC4uLg==",highlighted:`<span class="hljs-attr">policy_base_url:</span> <span class="hljs-string">http://127.0.0.1:8000/v1</span>
<span class="hljs-attr">policy_api_key:</span> <span class="hljs-string">EMPTY</span>
<span class="hljs-attr">policy_model_name:</span> <span class="hljs-string">Qwen/Qwen2.5-1.5B-Instruct</span>
<span class="hljs-attr">hf_token:</span> <span class="hljs-string">...</span>`,wrap:!1}}),Mt=new T({props:{title:"Run Training",local:"run-training",headingTag:"h3"}}),yt=new y({props:{code:"Y2QlMjBHeW0lMkYlMEFzb3VyY2UlMjAudmVudiUyRmJpbiUyRmFjdGl2YXRlJTBBJTBBY29uZmlnX3BhdGhzJTNEJTIycmVzb3VyY2VzX3NlcnZlcnMlMkZ3b3JrcGxhY2VfYXNzaXN0YW50JTJGY29uZmlncyUyRndvcmtwbGFjZV9hc3Npc3RhbnQueWFtbCUyQyU1QyUwQXJlc3BvbnNlc19hcGlfbW9kZWxzJTJGdmxsbV9tb2RlbCUyRmNvbmZpZ3MlMkZ2bGxtX21vZGVsX2Zvcl90cmFpbmluZy55YW1sJTIyJTBBJTBBbmdfcnVuJTIwJTIyJTJCY29uZmlnX3BhdGhzJTNEJTVCJTI0JTdCY29uZmlnX3BhdGhzJTdEJTVEJTIy",highlighted:`<span class="hljs-built_in">cd</span> Gym/
<span class="hljs-built_in">source</span> .venv/bin/activate

config_paths=<span class="hljs-string">&quot;resources_servers/workplace_assistant/configs/workplace_assistant.yaml,\\
responses_api_models/vllm_model/configs/vllm_model_for_training.yaml&quot;</span>

ng_run <span class="hljs-string">&quot;+config_paths=[<span class="hljs-variable">\${config_paths}</span>]&quot;</span>`,wrap:!1}}),Tt=new y({props:{code:"Y2QlMjB0cmwlMkYlMEFzb3VyY2UlMjAudmVudiUyRmJpbiUyRmFjdGl2YXRlJTBBQ1VEQV9WSVNJQkxFX0RFVklDRVMlM0QwJTIwdHJsJTIwdmxsbS1zZXJ2ZSUyMCU1QyUwQSUyMCUyMC0tbW9kZWwlMjBRd2VuJTJGUXdlbjIuNS0xLjVCLUluc3RydWN0JTIwJTVDJTBBJTIwJTIwLS1tYXgtbW9kZWwtbGVuJTIwMTYzODQlMjAlNUMlMEElMjAlMjAtLWhvc3QlMjAwLjAuMC4wJTIwJTVDJTBBJTIwJTIwLS1wb3J0JTIwODAwMA==",highlighted:`<span class="hljs-built_in">cd</span> trl/
<span class="hljs-built_in">source</span> .venv/bin/activate
CUDA_VISIBLE_DEVICES=0 trl vllm-serve \\
  --model Qwen/Qwen2.5-1.5B-Instruct \\
  --max-model-len 16384 \\
  --host 0.0.0.0 \\
  --port 8000`,wrap:!1}}),vt=new y({props:{code:"c291cmNlJTIwdHJsJTJGLnZlbnYlMkZiaW4lMkZhY3RpdmF0ZSUwQWNkJTIwdHJsJTJGZXhhbXBsZXMlMkZzY3JpcHRzJTJGbmVtb19neW0lMEFleHBvcnQlMjBXQU5EQl9BUElfS0VZJTNELi4uJTIwJTBBdXYlMjBhZGQlMjBvbWVnYWNvbmYlMjAlMEElMEFDVURBX1ZJU0lCTEVfREVWSUNFUyUzRDElMjBweXRob24lMjB0cmFpbl9tdWx0aV9lbnZpcm9ubWVudC5weSUyMC0tY29uZmlnJTIwY29uZmlnLnlhbWw=",highlighted:`<span class="hljs-built_in">source</span> trl/.venv/bin/activate
<span class="hljs-built_in">cd</span> trl/examples/scripts/nemo_gym
<span class="hljs-built_in">export</span> WANDB_API_KEY=... 
uv add omegaconf 

CUDA_VISIBLE_DEVICES=1 python train_multi_environment.py --config config.yaml`,wrap:!1}}),jt=new T({props:{title:"Multi-Node Training with Slurm",local:"multi-node-training-with-slurm",headingTag:"h2"}}),Ut=new y({props:{code:"c2JhdGNoJTIwc3VibWl0LnNo",highlighted:"sbatch submit.sh",wrap:!1}}),_t=new y({props:{code:"dGFpbCUyMC1mJTIwbG9ncyUyRiUzQ2pvYl9pZCUzRSUyRio=",highlighted:'<span class="hljs-built_in">tail</span> -f logs/&lt;job_id&gt;/*',wrap:!1}}),It=new T({props:{title:"Multi-Environment Training",local:"multi-environment-training",headingTag:"h2"}}),Lt=new y({props:{code:"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",highlighted:`<span class="hljs-built_in">cd</span> Gym
<span class="hljs-built_in">source</span> .venv/bin/activate
uv add reasoning-gym
<span class="hljs-built_in">cd</span> resources_servers/reasoning_gym
python scripts/create_dataset.py \\
    --task mini_sudoku \\
    --size 2000 \\
    --seed 42 \\
    --output data/reasoning_gym/train_mini_sudoku.jsonl

python scripts/create_dataset.py \\
    --task mini_sudoku \\
    --size 50 \\
    --seed 24 \\
    --output data/reasoning_gym/val_mini_sudoku.jsonl`,wrap:!1}}),Gt=new y({props:{code:"Y2F0JTIwZGF0YSUyRndvcmtwbGFjZV9hc3Npc3RhbnQlMkZ0cmFpbl93b3JrcGxhY2UuanNvbmwlMjBkYXRhJTJGcmVhc29uaW5nX2d5bSUyRnRyYWluX21pbmlfc3Vkb2t1Lmpzb25sJTIwJTdDJTIwc2h1ZiUyMCUzRSUyMHRyYWluX211bHRpX2Vudi5qc29ubA==",highlighted:'<span class="hljs-built_in">cat</span> data/workplace_assistant/train_workplace.jsonl data/reasoning_gym/train_mini_sudoku.jsonl | <span class="hljs-built_in">shuf</span> &gt; train_multi_env.jsonl',wrap:!1}}),Bt=new y({props:{code:"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",highlighted:`<span class="hljs-attr">model_name:</span> <span class="hljs-string">&quot;Qwen/Qwen3-4B-Instruct-2507&quot;</span>

<span class="hljs-attr">dataset_path:</span> <span class="hljs-string">&quot;/path/to/data/train_multi_env.jsonl&quot;</span>
<span class="hljs-attr">eval_dataset_path:</span> <span class="hljs-string">&quot;/path/to/data/val_multi_env.jsonl&quot;</span>

<span class="hljs-attr">task:</span> <span class="hljs-string">&quot;workplace-sudoku&quot;</span>                    <span class="hljs-comment"># used in wandb run name</span>
<span class="hljs-attr">output_dir:</span> <span class="hljs-string">&quot;outputs/nemo_gym_multi_env&quot;</span>

<span class="hljs-comment"># ... rest of config same</span>`,wrap:!1}}),Nt=new y({props:{code:"Y2QlMjBHeW0lMEFzb3VyY2UlMjAudmVudiUyRmJpbiUyRmFjdGl2YXRlJTBBJTBBY29uZmlnX3BhdGhzJTNEJTIycmVzcG9uc2VzX2FwaV9tb2RlbHMlMkZ2bGxtX21vZGVsJTJGY29uZmlncyUyRnZsbG1fbW9kZWwueWFtbCUyQyU1QyUwQXJlc291cmNlc19zZXJ2ZXJzJTJGd29ya3BsYWNlX2Fzc2lzdGFudCUyRmNvbmZpZ3MlMkZ3b3JrcGxhY2VfYXNzaXN0YW50LnlhbWwlMkMlNUMlMEFyZXNvdXJjZXNfc2VydmVycyUyRnJlYXNvbmluZ19neW0lMkZjb25maWdzJTJGcmVhc29uaW5nX2d5bS55YW1sJTIyJTBBJTBBbmdfcnVuJTIwJTIyJTJCY29uZmlnX3BhdGhzJTNEJTVCJTI0JTdCY29uZmlnX3BhdGhzJTdEJTVEJTIy",highlighted:`<span class="hljs-built_in">cd</span> Gym
<span class="hljs-built_in">source</span> .venv/bin/activate

config_paths=<span class="hljs-string">&quot;responses_api_models/vllm_model/configs/vllm_model.yaml,\\
resources_servers/workplace_assistant/configs/workplace_assistant.yaml,\\
resources_servers/reasoning_gym/configs/reasoning_gym.yaml&quot;</span>

ng_run <span class="hljs-string">&quot;+config_paths=[<span class="hljs-variable">\${config_paths}</span>]&quot;</span>`,wrap:!1}}),Et=new T({props:{title:"Resources",local:"resources",headingTag:"h2"}}),Xt=new sl({props:{source:"https://github.com/huggingface/trl/blob/main/docs/source/nemo_gym.md"}}),{c(){$=i("meta"),ve=l(),Te=i("p"),je=l(),u(Y.$$.fragment),be=l(),u(x.$$.fragment),we=l(),V=i("p"),V.textContent=ss,Ue=l(),k=i("p"),k.textContent=ls,$e=l(),u(H.$$.fragment),_e=l(),W=i("ul"),W.innerHTML=as,Ce=l(),u(S.$$.fragment),Ie=l(),A=i("p"),A.textContent=is,Ze=l(),F=i("table"),F.innerHTML=rs,Le=l(),Q=i("p"),Q.innerHTML=os,Ge=l(),u(q.$$.fragment),Be=l(),z=i("p"),z.textContent=ps,Ne=l(),u(P.$$.fragment),Ee=l(),G=i("ol"),D=i("li"),Yt=i("p"),Yt.innerHTML=us,Jn=l(),u(O.$$.fragment),vn=l(),K=i("li"),xt=i("p"),xt.innerHTML=ms,jn=l(),u(tt.$$.fragment),Re=l(),u(et.$$.fragment),Xe=l(),nt=i("p"),nt.innerHTML=cs,Ye=l(),st=i("ul"),st.innerHTML=ds,xe=l(),lt=i("blockquote"),lt.innerHTML=hs,Ve=l(),B=i("ol"),_=i("li"),Vt=i("p"),Vt.innerHTML=Ms,bn=l(),kt=i("p"),kt.innerHTML=gs,wn=l(),u(at.$$.fragment),Un=l(),C=i("li"),Ht=i("p"),Ht.innerHTML=ys,$n=l(),u(it.$$.fragment),_n=l(),Wt=i("p"),Wt.innerHTML=fs,ke=l(),rt=i("p"),rt.innerHTML=Ts,He=l(),u(ot.$$.fragment),We=l(),pt=i("p"),pt.textContent=Js,Se=l(),u(ut.$$.fragment),Ae=l(),u(mt.$$.fragment),Fe=l(),ct=i("p"),ct.textContent=vs,Qe=l(),u(dt.$$.fragment),qe=l(),N=i("ol"),I=i("li"),St=i("p"),St.innerHTML=js,Cn=l(),At=i("p"),At.innerHTML=bs,In=l(),u(ht.$$.fragment),Zn=l(),Ft=i("li"),Ft.innerHTML=ws,ze=l(),u(Mt.$$.fragment),Pe=l(),gt=i("p"),gt.textContent=Us,De=l(),b=i("ol"),J=i("li"),Qt=i("p"),Qt.innerHTML=$s,Ln=l(),u(yt.$$.fragment),Gn=l(),qt=i("p"),qt.textContent=_s,Bn=l(),zt=i("ul"),zt.innerHTML=Cs,Nn=l(),ft=i("li"),Pt=i("p"),Pt.innerHTML=Is,En=l(),u(Tt.$$.fragment),Rn=l(),Jt=i("li"),Dt=i("p"),Dt.innerHTML=Zs,Xn=l(),u(vt.$$.fragment),Oe=l(),u(jt.$$.fragment),Ke=l(),bt=i("p"),bt.innerHTML=Ls,tn=l(),w=i("ol"),Ot=i("li"),Ot.innerHTML=Gs,Yn=l(),wt=i("li"),Kt=i("p"),Kt.innerHTML=Bs,xn=l(),u(Ut.$$.fragment),Vn=l(),$t=i("li"),te=i("p"),te.innerHTML=Ns,kn=l(),u(_t.$$.fragment),en=l(),Ct=i("blockquote"),Ct.innerHTML=Es,nn=l(),u(It.$$.fragment),sn=l(),Zt=i("p"),Zt.textContent=Rs,ln=l(),f=i("ol"),Z=i("li"),ee=i("p"),ee.innerHTML=Xs,Hn=l(),ne=i("p"),ne.textContent=Ys,Wn=l(),u(Lt.$$.fragment),Sn=l(),v=i("li"),se=i("p"),se.innerHTML=xs,An=l(),le=i("p"),le.textContent=Vs,Fn=l(),u(Gt.$$.fragment),Qn=l(),ae=i("blockquote"),ae.innerHTML=ks,qn=l(),L=i("li"),ie=i("p"),ie.innerHTML=Hs,zn=l(),re=i("p"),re.textContent=Ws,Pn=l(),u(Bt.$$.fragment),Dn=l(),j=i("li"),oe=i("p"),oe.innerHTML=Ss,On=l(),pe=i("p"),pe.innerHTML=As,Kn=l(),u(Nt.$$.fragment),ts=l(),ue=i("p"),ue.innerHTML=Fs,es=l(),me=i("li"),me.innerHTML=Qs,an=l(),u(Et.$$.fragment),rn=l(),Rt=i("ul"),Rt.innerHTML=qs,on=l(),u(Xt.$$.fragment),pn=l(),Je=i("p"),this.h()},l(t){const n=el("svelte-u9bgzb",document.head);$=r(n,"META",{name:!0,content:!0}),n.forEach(e),ve=a(t),Te=r(t,"P",{}),g(Te).forEach(e),je=a(t),m(Y.$$.fragment,t),be=a(t),m(x.$$.fragment,t),we=a(t),V=r(t,"P",{"data-svelte-h":!0}),p(V)!=="svelte-7af6pp"&&(V.textContent=ss),Ue=a(t),k=r(t,"P",{"data-svelte-h":!0}),p(k)!=="svelte-iz4kji"&&(k.textContent=ls),$e=a(t),m(H.$$.fragment,t),_e=a(t),W=r(t,"UL",{"data-svelte-h":!0}),p(W)!=="svelte-1yst2fh"&&(W.innerHTML=as),Ce=a(t),m(S.$$.fragment,t),Ie=a(t),A=r(t,"P",{"data-svelte-h":!0}),p(A)!=="svelte-yszfde"&&(A.textContent=is),Ze=a(t),F=r(t,"TABLE",{"data-svelte-h":!0}),p(F)!=="svelte-13a2t9r"&&(F.innerHTML=rs),Le=a(t),Q=r(t,"P",{"data-svelte-h":!0}),p(Q)!=="svelte-1ah28g6"&&(Q.innerHTML=os),Ge=a(t),m(q.$$.fragment,t),Be=a(t),z=r(t,"P",{"data-svelte-h":!0}),p(z)!=="svelte-j2hll8"&&(z.textContent=ps),Ne=a(t),m(P.$$.fragment,t),Ee=a(t),G=r(t,"OL",{});var mn=g(G);D=r(mn,"LI",{});var cn=g(D);Yt=r(cn,"P",{"data-svelte-h":!0}),p(Yt)!=="svelte-1fllfb8"&&(Yt.innerHTML=us),Jn=a(cn),m(O.$$.fragment,cn),cn.forEach(e),vn=a(mn),K=r(mn,"LI",{});var dn=g(K);xt=r(dn,"P",{"data-svelte-h":!0}),p(xt)!=="svelte-wroi5s"&&(xt.innerHTML=ms),jn=a(dn),m(tt.$$.fragment,dn),dn.forEach(e),mn.forEach(e),Re=a(t),m(et.$$.fragment,t),Xe=a(t),nt=r(t,"P",{"data-svelte-h":!0}),p(nt)!=="svelte-xwoeod"&&(nt.innerHTML=cs),Ye=a(t),st=r(t,"UL",{"data-svelte-h":!0}),p(st)!=="svelte-1pj586x"&&(st.innerHTML=ds),xe=a(t),lt=r(t,"BLOCKQUOTE",{"data-svelte-h":!0}),p(lt)!=="svelte-prptho"&&(lt.innerHTML=hs),Ve=a(t),B=r(t,"OL",{});var hn=g(B);_=r(hn,"LI",{});var ce=g(_);Vt=r(ce,"P",{"data-svelte-h":!0}),p(Vt)!=="svelte-1ej3gnm"&&(Vt.innerHTML=Ms),bn=a(ce),kt=r(ce,"P",{"data-svelte-h":!0}),p(kt)!=="svelte-dolnl9"&&(kt.innerHTML=gs),wn=a(ce),m(at.$$.fragment,ce),ce.forEach(e),Un=a(hn),C=r(hn,"LI",{});var de=g(C);Ht=r(de,"P",{"data-svelte-h":!0}),p(Ht)!=="svelte-1t97quy"&&(Ht.innerHTML=ys),$n=a(de),m(it.$$.fragment,de),_n=a(de),Wt=r(de,"P",{"data-svelte-h":!0}),p(Wt)!=="svelte-1a2biir"&&(Wt.innerHTML=fs),de.forEach(e),hn.forEach(e),ke=a(t),rt=r(t,"P",{"data-svelte-h":!0}),p(rt)!=="svelte-fceadt"&&(rt.innerHTML=Ts),He=a(t),m(ot.$$.fragment,t),We=a(t),pt=r(t,"P",{"data-svelte-h":!0}),p(pt)!=="svelte-1f9su50"&&(pt.textContent=Js),Se=a(t),m(ut.$$.fragment,t),Ae=a(t),m(mt.$$.fragment,t),Fe=a(t),ct=r(t,"P",{"data-svelte-h":!0}),p(ct)!=="svelte-doprer"&&(ct.textContent=vs),Qe=a(t),m(dt.$$.fragment,t),qe=a(t),N=r(t,"OL",{});var Mn=g(N);I=r(Mn,"LI",{});var he=g(I);St=r(he,"P",{"data-svelte-h":!0}),p(St)!=="svelte-1jza2zv"&&(St.innerHTML=js),Cn=a(he),At=r(he,"P",{"data-svelte-h":!0}),p(At)!=="svelte-5yzoe6"&&(At.innerHTML=bs),In=a(he),m(ht.$$.fragment,he),he.forEach(e),Zn=a(Mn),Ft=r(Mn,"LI",{"data-svelte-h":!0}),p(Ft)!=="svelte-18xbdkv"&&(Ft.innerHTML=ws),Mn.forEach(e),ze=a(t),m(Mt.$$.fragment,t),Pe=a(t),gt=r(t,"P",{"data-svelte-h":!0}),p(gt)!=="svelte-1ji7mz"&&(gt.textContent=Us),De=a(t),b=r(t,"OL",{});var Me=g(b);J=r(Me,"LI",{});var E=g(J);Qt=r(E,"P",{"data-svelte-h":!0}),p(Qt)!=="svelte-1i1pb51"&&(Qt.innerHTML=$s),Ln=a(E),m(yt.$$.fragment,E),Gn=a(E),qt=r(E,"P",{"data-svelte-h":!0}),p(qt)!=="svelte-1nv1hcv"&&(qt.textContent=_s),Bn=a(E),zt=r(E,"UL",{"data-svelte-h":!0}),p(zt)!=="svelte-1dlzpa1"&&(zt.innerHTML=Cs),E.forEach(e),Nn=a(Me),ft=r(Me,"LI",{});var gn=g(ft);Pt=r(gn,"P",{"data-svelte-h":!0}),p(Pt)!=="svelte-ntp91"&&(Pt.innerHTML=Is),En=a(gn),m(Tt.$$.fragment,gn),gn.forEach(e),Rn=a(Me),Jt=r(Me,"LI",{});var yn=g(Jt);Dt=r(yn,"P",{"data-svelte-h":!0}),p(Dt)!=="svelte-1d1ejwe"&&(Dt.innerHTML=Zs),Xn=a(yn),m(vt.$$.fragment,yn),yn.forEach(e),Me.forEach(e),Oe=a(t),m(jt.$$.fragment,t),Ke=a(t),bt=r(t,"P",{"data-svelte-h":!0}),p(bt)!=="svelte-h26aax"&&(bt.innerHTML=Ls),tn=a(t),w=r(t,"OL",{});var ge=g(w);Ot=r(ge,"LI",{"data-svelte-h":!0}),p(Ot)!=="svelte-skqx2a"&&(Ot.innerHTML=Gs),Yn=a(ge),wt=r(ge,"LI",{});var fn=g(wt);Kt=r(fn,"P",{"data-svelte-h":!0}),p(Kt)!=="svelte-13q6911"&&(Kt.innerHTML=Bs),xn=a(fn),m(Ut.$$.fragment,fn),fn.forEach(e),Vn=a(ge),$t=r(ge,"LI",{});var Tn=g($t);te=r(Tn,"P",{"data-svelte-h":!0}),p(te)!=="svelte-i5tzjh"&&(te.innerHTML=Ns),kn=a(Tn),m(_t.$$.fragment,Tn),Tn.forEach(e),ge.forEach(e),en=a(t),Ct=r(t,"BLOCKQUOTE",{"data-svelte-h":!0}),p(Ct)!=="svelte-14ze16n"&&(Ct.innerHTML=Es),nn=a(t),m(It.$$.fragment,t),sn=a(t),Zt=r(t,"P",{"data-svelte-h":!0}),p(Zt)!=="svelte-b0bbwm"&&(Zt.textContent=Rs),ln=a(t),f=r(t,"OL",{});var U=g(f);Z=r(U,"LI",{});var ye=g(Z);ee=r(ye,"P",{"data-svelte-h":!0}),p(ee)!=="svelte-1upr22m"&&(ee.innerHTML=Xs),Hn=a(ye),ne=r(ye,"P",{"data-svelte-h":!0}),p(ne)!=="svelte-14v8vct"&&(ne.textContent=Ys),Wn=a(ye),m(Lt.$$.fragment,ye),ye.forEach(e),Sn=a(U),v=r(U,"LI",{});var R=g(v);se=r(R,"P",{"data-svelte-h":!0}),p(se)!=="svelte-1w7geiq"&&(se.innerHTML=xs),An=a(R),le=r(R,"P",{"data-svelte-h":!0}),p(le)!=="svelte-1atlcqa"&&(le.textContent=Vs),Fn=a(R),m(Gt.$$.fragment,R),Qn=a(R),ae=r(R,"BLOCKQUOTE",{"data-svelte-h":!0}),p(ae)!=="svelte-3gj9yl"&&(ae.innerHTML=ks),R.forEach(e),qn=a(U),L=r(U,"LI",{});var fe=g(L);ie=r(fe,"P",{"data-svelte-h":!0}),p(ie)!=="svelte-1r6uj0q"&&(ie.innerHTML=Hs),zn=a(fe),re=r(fe,"P",{"data-svelte-h":!0}),p(re)!=="svelte-9zp78s"&&(re.textContent=Ws),Pn=a(fe),m(Bt.$$.fragment,fe),fe.forEach(e),Dn=a(U),j=r(U,"LI",{});var X=g(j);oe=r(X,"P",{"data-svelte-h":!0}),p(oe)!=="svelte-1jr36l7"&&(oe.innerHTML=Ss),On=a(X),pe=r(X,"P",{"data-svelte-h":!0}),p(pe)!=="svelte-1l0rsb0"&&(pe.innerHTML=As),Kn=a(X),m(Nt.$$.fragment,X),ts=a(X),ue=r(X,"P",{"data-svelte-h":!0}),p(ue)!=="svelte-1e41fqn"&&(ue.innerHTML=Fs),X.forEach(e),es=a(U),me=r(U,"LI",{"data-svelte-h":!0}),p(me)!=="svelte-21uz4m"&&(me.innerHTML=Qs),U.forEach(e),an=a(t),m(Et.$$.fragment,t),rn=a(t),Rt=r(t,"UL",{"data-svelte-h":!0}),p(Rt)!=="svelte-3m8o0j"&&(Rt.innerHTML=qs),on=a(t),m(Xt.$$.fragment,t),pn=a(t),Je=r(t,"P",{}),g(Je).forEach(e),this.h()},h(){zs($,"name","hf:doc:metadata"),zs($,"content",al)},m(t,n){s(document.head,$),o(t,ve,n),o(t,Te,n),o(t,je,n),c(Y,t,n),o(t,be,n),c(x,t,n),o(t,we,n),o(t,V,n),o(t,Ue,n),o(t,k,n),o(t,$e,n),c(H,t,n),o(t,_e,n),o(t,W,n),o(t,Ce,n),c(S,t,n),o(t,Ie,n),o(t,A,n),o(t,Ze,n),o(t,F,n),o(t,Le,n),o(t,Q,n),o(t,Ge,n),c(q,t,n),o(t,Be,n),o(t,z,n),o(t,Ne,n),c(P,t,n),o(t,Ee,n),o(t,G,n),s(G,D),s(D,Yt),s(D,Jn),c(O,D,null),s(G,vn),s(G,K),s(K,xt),s(K,jn),c(tt,K,null),o(t,Re,n),c(et,t,n),o(t,Xe,n),o(t,nt,n),o(t,Ye,n),o(t,st,n),o(t,xe,n),o(t,lt,n),o(t,Ve,n),o(t,B,n),s(B,_),s(_,Vt),s(_,bn),s(_,kt),s(_,wn),c(at,_,null),s(B,Un),s(B,C),s(C,Ht),s(C,$n),c(it,C,null),s(C,_n),s(C,Wt),o(t,ke,n),o(t,rt,n),o(t,He,n),c(ot,t,n),o(t,We,n),o(t,pt,n),o(t,Se,n),c(ut,t,n),o(t,Ae,n),c(mt,t,n),o(t,Fe,n),o(t,ct,n),o(t,Qe,n),c(dt,t,n),o(t,qe,n),o(t,N,n),s(N,I),s(I,St),s(I,Cn),s(I,At),s(I,In),c(ht,I,null),s(N,Zn),s(N,Ft),o(t,ze,n),c(Mt,t,n),o(t,Pe,n),o(t,gt,n),o(t,De,n),o(t,b,n),s(b,J),s(J,Qt),s(J,Ln),c(yt,J,null),s(J,Gn),s(J,qt),s(J,Bn),s(J,zt),s(b,Nn),s(b,ft),s(ft,Pt),s(ft,En),c(Tt,ft,null),s(b,Rn),s(b,Jt),s(Jt,Dt),s(Jt,Xn),c(vt,Jt,null),o(t,Oe,n),c(jt,t,n),o(t,Ke,n),o(t,bt,n),o(t,tn,n),o(t,w,n),s(w,Ot),s(w,Yn),s(w,wt),s(wt,Kt),s(wt,xn),c(Ut,wt,null),s(w,Vn),s(w,$t),s($t,te),s($t,kn),c(_t,$t,null),o(t,en,n),o(t,Ct,n),o(t,nn,n),c(It,t,n),o(t,sn,n),o(t,Zt,n),o(t,ln,n),o(t,f,n),s(f,Z),s(Z,ee),s(Z,Hn),s(Z,ne),s(Z,Wn),c(Lt,Z,null),s(f,Sn),s(f,v),s(v,se),s(v,An),s(v,le),s(v,Fn),c(Gt,v,null),s(v,Qn),s(v,ae),s(f,qn),s(f,L),s(L,ie),s(L,zn),s(L,re),s(L,Pn),c(Bt,L,null),s(f,Dn),s(f,j),s(j,oe),s(j,On),s(j,pe),s(j,Kn),c(Nt,j,null),s(j,ts),s(j,ue),s(f,es),s(f,me),o(t,an,n),c(Et,t,n),o(t,rn,n),o(t,Rt,n),o(t,on,n),c(Xt,t,n),o(t,pn,n),o(t,Je,n),un=!0},p:Ds,i(t){un||(d(Y.$$.fragment,t),d(x.$$.fragment,t),d(H.$$.fragment,t),d(S.$$.fragment,t),d(q.$$.fragment,t),d(P.$$.fragment,t),d(O.$$.fragment,t),d(tt.$$.fragment,t),d(et.$$.fragment,t),d(at.$$.fragment,t),d(it.$$.fragment,t),d(ot.$$.fragment,t),d(ut.$$.fragment,t),d(mt.$$.fragment,t),d(dt.$$.fragment,t),d(ht.$$.fragment,t),d(Mt.$$.fragment,t),d(yt.$$.fragment,t),d(Tt.$$.fragment,t),d(vt.$$.fragment,t),d(jt.$$.fragment,t),d(Ut.$$.fragment,t),d(_t.$$.fragment,t),d(It.$$.fragment,t),d(Lt.$$.fragment,t),d(Gt.$$.fragment,t),d(Bt.$$.fragment,t),d(Nt.$$.fragment,t),d(Et.$$.fragment,t),d(Xt.$$.fragment,t),un=!0)},o(t){h(Y.$$.fragment,t),h(x.$$.fragment,t),h(H.$$.fragment,t),h(S.$$.fragment,t),h(q.$$.fragment,t),h(P.$$.fragment,t),h(O.$$.fragment,t),h(tt.$$.fragment,t),h(et.$$.fragment,t),h(at.$$.fragment,t),h(it.$$.fragment,t),h(ot.$$.fragment,t),h(ut.$$.fragment,t),h(mt.$$.fragment,t),h(dt.$$.fragment,t),h(ht.$$.fragment,t),h(Mt.$$.fragment,t),h(yt.$$.fragment,t),h(Tt.$$.fragment,t),h(vt.$$.fragment,t),h(jt.$$.fragment,t),h(Ut.$$.fragment,t),h(_t.$$.fragment,t),h(It.$$.fragment,t),h(Lt.$$.fragment,t),h(Gt.$$.fragment,t),h(Bt.$$.fragment,t),h(Nt.$$.fragment,t),h(Et.$$.fragment,t),h(Xt.$$.fragment,t),un=!1},d(t){t&&(e(ve),e(Te),e(je),e(be),e(we),e(V),e(Ue),e(k),e($e),e(_e),e(W),e(Ce),e(Ie),e(A),e(Ze),e(F),e(Le),e(Q),e(Ge),e(Be),e(z),e(Ne),e(Ee),e(G),e(Re),e(Xe),e(nt),e(Ye),e(st),e(xe),e(lt),e(Ve),e(B),e(ke),e(rt),e(He),e(We),e(pt),e(Se),e(Ae),e(Fe),e(ct),e(Qe),e(qe),e(N),e(ze),e(Pe),e(gt),e(De),e(b),e(Oe),e(Ke),e(bt),e(tn),e(w),e(en),e(Ct),e(nn),e(sn),e(Zt),e(ln),e(f),e(an),e(rn),e(Rt),e(on),e(pn),e(Je)),e($),M(Y,t),M(x,t),M(H,t),M(S,t),M(q,t),M(P,t),M(O),M(tt),M(et,t),M(at),M(it),M(ot,t),M(ut,t),M(mt,t),M(dt,t),M(ht),M(Mt,t),M(yt),M(Tt),M(vt),M(jt,t),M(Ut),M(_t),M(It,t),M(Lt),M(Gt),M(Bt),M(Nt),M(Et,t),M(Xt,t)}}}const al='{"title":"NeMo Gym Integration","local":"nemo-gym-integration","sections":[{"title":"Why NeMo Gym","local":"why-nemo-gym","sections":[],"depth":2},{"title":"Available Environments","local":"available-environments","sections":[],"depth":2},{"title":"Before You Start","local":"before-you-start","sections":[{"title":"Install TRL and NeMo Gym","local":"install-trl-and-nemo-gym","sections":[],"depth":3},{"title":"Prepare a Dataset","local":"prepare-a-dataset","sections":[{"title":"Dataset Format","local":"dataset-format","sections":[],"depth":4}],"depth":3}],"depth":2},{"title":"Interactive Training","local":"interactive-training","sections":[{"title":"Set Up","local":"set-up","sections":[],"depth":3},{"title":"Run Training","local":"run-training","sections":[],"depth":3}],"depth":2},{"title":"Multi-Node Training with Slurm","local":"multi-node-training-with-slurm","sections":[],"depth":2},{"title":"Multi-Environment Training","local":"multi-environment-training","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2}],"depth":1}';function il(ns){return Os(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class ml extends Ks{constructor($){super(),tl(this,$,il,ll,Ps,{})}}export{ml as component};