openenv/pr_1049/en/tutorials/pi-agent-grpo.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Coding Agent Training with TRL (Pi)&quot;,&quot;local&quot;:&quot;coding-agent-training-with-trl-pi&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;How It Works&quot;,&quot;local&quot;:&quot;how-it-works&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Full Recipe&quot;,&quot;local&quot;:&quot;full-recipe&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/openenv/pr_1049/en/_app/immutable/entry/start.MXLlDTbg.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/CRNC0_Gt.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/BSuxAqoA.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/entry/app.B1F6BRmw.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/Ds_1Ru_R.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/CMopE-k8.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/BAUKCucN.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/nodes/0.XJht84dh.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/CR5HsSXT.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/nodes/66.CXkgWaYp.js" rel="modulepreload">
		<link href="/docs/openenv/pr_1049/en/_app/immutable/chunks/C07d6Hje.js" rel="modulepreload">
		<!--ovm1px--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Coding Agent Training with TRL (Pi)&quot;,&quot;local&quot;:&quot;coding-agent-training-with-trl-pi&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;How It Works&quot;,&quot;local&quot;:&quot;how-it-works&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Full Recipe&quot;,&quot;local&quot;:&quot;full-recipe&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/openenv/pr_1049/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><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> <!--[-1--><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="coding-agent-training-with-trl-pi" 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="#coding-agent-training-with-trl-pi"><span><svg 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>Coding Agent Training with TRL (Pi)</span></h1><!--]--><!----> <p>This tutorial covers the black-box training path: training the actual <a href="https://github.com/badlogic/pi-mono" rel="nofollow"><code>pi</code></a> coding agent, with its own planner,
tools, context management, and stop condition, using TRL’s experimental <code>AsyncGRPOTrainer</code>. The agent owns its loop, and OpenEnv captures what it did.</p> <blockquote class="note"><p>Three GRPO patterns, three tutorials. For a standard <code>reset()</code> / <code>step()</code> flow where TRL drives the episode, see the <a href="wordle-grpo">Wordle GRPO tutorial</a>. For harness rollouts where the
trainer still generates each turn (white-box), see the <a href="browsergym-harness">BrowserGym harness tutorial</a>. Use this page when you
want to train a production agent as-is, without reimplementing its loop.</p></blockquote> <!--[1--><h2 class="relative group"><a id="how-it-works" 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="#how-it-works"><span><svg 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>How It Works</span></h2><!--]--><!----> <p>The full recipe lives in TRL. The moving pieces:</p> <ol><li>Each rollout runs the agent inside an OpenEnv session created by <code>PiSessionFactory</code> from <a href="https://github.com/huggingface/OpenEnv/tree/main/envs/pi_env" rel="nofollow"><code>pi_env</code></a>,
in <code>transparent_proxy</code> mode. A small proxy inside the sandbox forwards the
agent’s <code>/v1/chat/completions</code> calls to your vLLM server and records each
turn’s token ids and logprobs to a trace.</li> <li>When the agent stops, TRL’s <code>HarnessRolloutWorker</code> reads the trace, rebuilds
the per-turn training rows from the recorded ids, and scores the final
workspace with the session’s <code>verify()</code> method (a held-out verifier the
agent never sees).</li> <li><code>AsyncGRPOTrainer</code> trains on those rows, propagating the rollout reward to
every trained token through the group-relative advantage. NCCL weight sync
keeps the vLLM server on the current policy, so the agent always samples
from the model being trained.</li></ol> <p>Each rollout gets its own isolated session: one sandbox, one proxy port, one
agent process. Three small functions adapt the recipe to your task: <code>rollout_reward_fn</code> (outcome to scalar reward), <code>train_turn_fn</code> (which turns
receive gradient), and <code>agent_turn_fn</code> (which trace entries are real agent
turns rather than auxiliary calls like title generation).</p> <!--[1--><h2 class="relative group"><a id="full-recipe" 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="#full-recipe"><span><svg 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>Full Recipe</span></h2><!--]--><!----> <p>The reference script trains on competitive-coding problems from <code>agentica-org/DeepCoder-Preview-Dataset</code>: the agent writes <code>solution.py</code>, and
the verifier runs it against held-out tests, returning the fraction passed. It
is self-contained, runs the agent in a local subprocess sandbox (no container
setup needed), needs two GPUs (one serving the policy with vLLM, one
training).</p> <ul><li><a href="https://github.com/huggingface/trl/blob/main/examples/scripts/openenv/pi.py" rel="nofollow"><code>examples/scripts/openenv/pi.py</code></a> in TRL: the complete, runnable script.</li> <li><a href="https://github.com/huggingface/OpenEnv/tree/main/envs/pi_env" rel="nofollow"><code>envs/pi_env</code></a>:
the OpenEnv side, including the session factory, sandbox backends, and the
transparent interception proxy.</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/openenv/blob/main/docs/source/tutorials/pi-agent-grpo.md" 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><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->
			
			<script>
				{
					__sveltekit_1mo3g52 = {
						base: "/docs/openenv/pr_1049/en",
						assets: "/docs/openenv/pr_1049/en"
					};

					const element = document.currentScript.parentElement;

					Promise.all([
						import("/docs/openenv/pr_1049/en/_app/immutable/entry/start.MXLlDTbg.js"),
						import("/docs/openenv/pr_1049/en/_app/immutable/entry/app.B1F6BRmw.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 66],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>