<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>verl Blog</title>
		<link>https://verl-project.github.io/</link>
		<description>Recent content on verl Blog</description>
		<generator>Hugo</generator>
		<language>en-us</language>
		
		
		
			<copyright>verl-project</copyright>
		
		
			<lastBuildDate>Mon, 16 Nov 2026 00:00:00 +0000</lastBuildDate>
		
			<atom:link href="https://verl-project.github.io/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>verl-SpeCo v0.2.0: Faster Standalone Training and Broader Online Co-Training</title>
				<link>https://verl-project.github.io/posts/2026-11-16-verl-speco-v0-2-0/</link>
				<pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-11-16-verl-speco-v0-2-0/</guid>
				<description>verl-SpeCo v0.2.0 pipelines standalone drafter training, extends online DSpark co-training to V1, OPD, and SGLang, and adds controls for when and on what to train.</description>
			</item>
			<item>
				<title>DeepSeek-V4-Flash RL on AMD MI355X with verl: 300 Stable Steps and FP8 Rollout Alignment</title>
				<link>https://verl-project.github.io/posts/2026-10-30-deepseek-v4-flash-rl-rocm/</link>
				<pubDate>Fri, 30 Oct 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-10-30-deepseek-v4-flash-rl-rocm/</guid>
				<description>verl runs DeepSeek-V4-Flash GRPO on 32 AMD Instinct MI355X GPUs: training aligned with FP8 rollout quantization, and BF16, BF16+QAT and FP8 recipes each completed 300 stable steps.</description>
			</item>
			<item>
				<title>VeRL-Omni v0.3.0: Async Training and Broader Multimodal Post-Training</title>
				<link>https://verl-project.github.io/posts/2026-10-29-verl-omni-v0-3-0/</link>
				<pubDate>Thu, 29 Oct 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-10-29-verl-omni-v0-3-0/</guid>
				<description>VeRL-Omni v0.3.0 expands async RL, makes the V1 trainer the default, and adds reproducible recipes for audio-video generation and multimodal understanding.</description>
			</item>
			<item>
				<title>Async RL Training Tuning Guide</title>
				<link>https://verl-project.github.io/posts/2026-10-23-async-rl-training-tuning/</link>
				<pubDate>Mon, 26 Oct 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-10-23-async-rl-training-tuning/</guid>
				<description>Use RL Insight to locate Trainer and Rollouter bubbles, properly balance training and inference resources, maximize single-card throughput, and enable more efficient training.</description>
			</item>
			<item>
				<title>End-to-End MXFP8 RL on Blackwell: What Matching Buys, When It Pays Off, and How It Fails</title>
				<link>https://verl-project.github.io/posts/2026-10-23-mxfp8-end-to-end-rl/</link>
				<pubDate>Fri, 23 Oct 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-10-23-mxfp8-end-to-end-rl/</guid>
				<description>MXFP8 on both sides of RL in verl on B200: a sampler on the learner&amp;rsquo;s weight grid cuts train-rollout KL by 16%, training pays off above about 4.5k tokens per micro-batch, and seven failures show what a sync must preserve.</description>
			</item>
			<item>
				<title>Uni-Agent: Train Long-Horizon Agents at Scale</title>
				<link>https://verl-project.github.io/posts/2026-09-23-uni-agent-v0-1-0/</link>
				<pubDate>Thu, 24 Sep 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-09-23-uni-agent-v0-1-0/</guid>
				<description>Uni-Agent is an open framework for scalable long-horizon agent RL. We share our insights on harness adaptation, RL training, and practical recipes for agentic RL.</description>
			</item>
			<item>
				<title>verl-vla v0.1.0: A Unified Cloud-Edge Post-Training Loop for Large-Scale VLA</title>
				<link>https://verl-project.github.io/posts/2026-09-02-verl-vla-v0-1-0/</link>
				<pubDate>Wed, 09 Sep 2026 23:59:00 +0800</pubDate>
				<guid>https://verl-project.github.io/posts/2026-09-02-verl-vla-v0-1-0/</guid>
				<description>verl-vla is an open-source framework built on verl that unifies human-in-the-loop data collection, fine-tuning, evaluation, and reinforcement learning for vision-language-action policies across cloud and edge resources.</description>
			</item>
			<item>
				<title>Bringing End-to-End LLM Reinforcement Learning to AMD Instinct MI455X with verl</title>
				<link>https://verl-project.github.io/posts/2026-09-09-verl-mi455x-bringup/</link>
				<pubDate>Wed, 09 Sep 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-09-09-verl-mi455x-bringup/</guid>
				<description>A complete GRPO workflow using verl, Qwen3-30B-A3B, SGLang, and PyTorch FSDP ran for 100 consecutive steps on four AMD Instinct MI455X GPUs, establishing an end-to-end RL baseline for CDNA 5.</description>
			</item>
			<item>
				<title>VeRL-Omni v0.2.0: Faster Diffusion RL and Stable Omni Training</title>
				<link>https://verl-project.github.io/posts/2026-08-17-verl-omni-v0-2-0/</link>
				<pubDate>Mon, 17 Aug 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-08-17-verl-omni-v0-2-0/</guid>
				<description>A release focused on higher-throughput diffusion rollout, reusable omni adapters, and broader recipe coverage.</description>
			</item>
			<item>
				<title>verl on AMD: production-ready RL post-training on ROCm</title>
				<link>https://verl-project.github.io/posts/2026-08-14-verl-on-amd-rocm/</link>
				<pubDate>Fri, 14 Aug 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-08-14-verl-on-amd-rocm/</guid>
				<description>RL post-training on AMD Instinct GPUs is here: a turnkey ROCm container, AITER-accelerated vLLM and SGLang rollout, and accuracy validated on both MI300 and MI350 series.</description>
			</item>
			<item>
				<title>Faster Draft Model Training: verl-SpeCo 0.1.0 Adds Native DSpark Support and Unlocks Standalone Training</title>
				<link>https://verl-project.github.io/posts/2026-08-13-verl-speco-release/</link>
				<pubDate>Thu, 13 Aug 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-08-13-verl-speco-release/</guid>
				<description>verl-SpeCo 0.1.0 adds native DSpark support and standalone draft model training, with reusable feature storage, seven algorithm backends, vLLM/SGLang integration, and GPU/Ascend NPU support.</description>
			</item>
			<item>
				<title>Introducing verl-tinker: Your Tinker loop, your GPUs</title>
				<link>https://verl-project.github.io/posts/2026-08-11-verl-tinker/</link>
				<pubDate>Tue, 11 Aug 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-08-11-verl-tinker/</guid>
				<description>Keep the Tinker Cookbook loop you know, and run SFT, RL, and distillation on verl-managed GPU workers you control.</description>
			</item>
			<item>
				<title>Introducing the verl blog</title>
				<link>https://verl-project.github.io/posts/2026-08-04-introducing-the-verl-blog/</link>
				<pubDate>Tue, 04 Aug 2026 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/posts/2026-08-04-introducing-the-verl-blog/</guid>
				<description>verl now has its own blog. It exists so that the people who build features can explain them properly.</description>
			</item>
			<item>
				<title>Write for the verl blog</title>
				<link>https://verl-project.github.io/contribute/</link>
				<pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate>
				<guid>https://verl-project.github.io/contribute/</guid>
				<description>How to propose, write and ship a post about work you have done in verl.</description>
			</item>
	</channel>
</rss>
