Think Swift, Build More: Empowering Agents with RL Toolkits and Infrastructure
[In-person event in San Francisco] [50 spots only]
This event is cohosted by verl, SGLang, Zilliz and Creao AI and organized by Monolith Management. Together, we’ll explore the latest advances in RL, RL infrastructure, Reasoning, and Agentic AI
We’ll open with several presentations and dig into:
verl – Reinforcement Learning framework designed for efficient and flexible training of large-scale models
Optimizing End2End Multi-turn RL with SGLang rollout
Tool uses a feature on SGLang with various tool parsers
SpecForge: A unified training framework for speculative decoding across LLMs, VLMs, and LoRAs
Zilliz – Unlocking billion-scale AI search with Milvus for massive unstructured data
Creao AI – Building tools and infrastructure for code agent
After the presentations, stick around for drinks, snacks, and deep-cut conversations with other engineers and researchers from OpenAI, Anthropic, MSL, xAI, and more
About Co-hosts & Speakers
verl: An open-source RL framework for LLMs, enabling efficient training across large-scale distributed systems with 300+ contributors and 12k+ stars on Github
Haibin Lin works on foundation models at Bytedance Seed, focusing on optimizing training infra for LLMs & multimodal models (with more than 10k GPUs), from pre-training to post-training. Prior to Bytedance, he works on Apache MXNet.
Ziheng Jiang works on LLM systems at Bytedance, focusing on scaling and optimizing LLM training and inference. His research work spans projects such as MegaScale (with 10K+ GPUs), MegaScale-Infer and Apache TVM.
SGLang: An open-source, fast serving and programming framework for LLMs and VLMs with 600+ contributors and 17k+ stars on Github
Chenyang Zhao, a Ph.D. student in Computer Science at UCLA. A proud member of the lmsys.org community and the leader of the SGLang RL team. Focus on building and optimizing Post-training Pipelines and Agentic RL Systems
Shuai Shi， Lead ML engineer in Cloudsway.AI, SpecForge maintainer, SGLang Core contributor, focus on MLLM inference and RL
Zilliz / Milvus: An open-source high-performance vector database powering large-scale AI search across text, images, and multi-modal data with 300+ contributors and 36k+ stars on Github
Jiang Chen is the Head of Developer Relations at Zilliz, the creator of Milvus open-source vector database. Previously, he was a tech lead and product manager at Google, leading the development of web-scale semantic understanding and search indexing
Creao AI: Building the foundational agentic infrastructure for a world where autonomous agents actively work on our behalf
Peter Pang, Co-founder and CTO of Creao AI, former Apple AI Scientist and Meta GenAI tech lead, is now pioneering AI-native platforms at his startup, advancing the frontier of agentic AI where humans and agents work side by side.
About Organizer
Monolith Management takes its name from the enigmatic black monolith in the iconic sci-fi movie "2001: A Space odyssey.", symbolizing universal truths with its 1:4:9 ratio—the squares of the first three natural numbers. At Monolith Management, we embody curiosity, resilience, and wisdom, blending detached thinking with decisive action. We aim to stand shoulder-to-shoulder with entrepreneurs, merging visionary concepts with steadfast execution to jointly shape the future.
M-A-P is an open-source AI research community.The community members are working on research topics in a wide range of spectrum, including but not limited to pre-training paradigm of foundation models, large-scale data collection and processing, and the derived applciations on coding, reasoning and music creativity.
Who should attend:
For researchers, developers, technical builders and founders working with Agentic AI, Coding, Reasoning or RL Infrastructure
Limited capacity. RSVP required.
This event is on-site only, no recording available. If you cannot make it, please cancel registration