Staff AI Engineer, Model Post-Training and Alignment

职位要求 / 描述

<h2 class="heading-2 ace-line old-record-id-doxusd6NtLTrM8e4ysmP03OKuHd"><strong>Who We Are</strong></h2> <p>At OKX, we believe that the future will be reshaped by crypto, and ultimately contribute to every individual's freedom. OKX is a leading crypto exchange, and the developer of OKX Wallet, giving millions access to crypto trading and decentralized crypto applications (dApps). OKX is also a trusted brand by hundreds of large institutions seeking access to crypto markets. We are safe and reliable, backed by our Proof of Reserves. Across our multiple offices globally, we are united by our core principles: <em>We Before Me</em>, <em>Do the Right Thing</em>, and <em>Get Things Done</em>. These shared values drive our culture, shape our processes, and foster a friendly, rewarding, and diverse environment for every OK-er. OKX is part of OKG, a group that brings the value of Blockchain to users around the world, through our leading products OKX, OKX Wallet, OKLink and more.</p> <h2 class="heading-2 ace-line old-record-id-doxusl417D129Ked7hp9WBVv9lg"><strong>About the Opportunity</strong></h2> <div data-page-id="YnvfdojQpornTcxhh7FlZ37ugze" data-lark-html-role="root" data-docx-has-block-data="false"> <div class="ace-line ace-line old-record-id-Foo4dvQATotKkuxbUEglUO6sg2J"> <p data-start="176" data-end="486">We are seeking a highly skilled and hands-on Machine Learning Engineer specializing in <strong>large model post-training and alignment</strong>. This role focuses on designing, executing, and optimizing post-training pipelines to improve model performance, controllability, domain adaptation, and reasoning capabilities.</p> <p data-start="488" data-end="671">You will work across the full lifecycle of post-training—from data strategy and reward modeling to reinforcement learning–based optimization and production-grade inference deployment.</p> <p data-start="488" data-end="671"> </p> </div> </div> <h2 class="heading-2 ace-line old-record-id-doxusI2RkaIWtuQsGjyJjA6hybg"><strong>What You’ll Be Doing </strong></h2> <div data-page-id="YnvfdojQpornTcxhh7FlZ37ugze" data-lark-html-role="root" data-docx-has-block-data="false"> <ul class="list-bullet1"> <li class="ace-line ace-line old-record-id-JLKOdU6cwoyIjbxLKZxlEttugxc" data-list="bullet">Lead and execute the full post-training pipeline for large language models (LLMs), including supervised fine-tuning, preference optimization, and reinforcement learning–based methods.</li> <li class="ace-line ace-line old-record-id-AFvSdoB44oXpewxukzelPublgCe" data-list="bullet">Design and implement advanced training paradigms such as <strong data-start="947" data-end="987">DPO (Direct Preference Optimization)</strong> and <strong data-start="992" data-end="1041">GRPO (Generalized Reward Policy Optimization)</strong>.</li> <li class="ace-line ace-line old-record-id-JXHrdNTcUoZBvWxvj1HlVENig3d" data-list="bullet">Develop domain-specific data recipes, curation strategies, and augmentation pipelines to optimize task performance.</li> <li class="ace-line ace-line old-record-id-WnzhdfL28oLSntxM2KilxZrBgCg" data-list="bullet">Conduct post-training of specialized small models from scratch, including architecture selection, dataset construction, and optimization strategy.</li> <li class="ace-line ace-line old-record-id-PKU2dkSMhoPMEuxuUiclXKKAgmg" data-list="bullet">Build and refine <strong data-start="1329" data-end="1346">Reward Models</strong> to support alignment and downstream optimization.</li> <li class="ace-line ace-line old-record-id-PKU2dkSMhoPMEuxuUiclXKKAgmg" data-list="bullet">Design and implement <strong data-start="1420" data-end="1471">RLAIF (Reinforcement Learning from AI Feedback)</strong> closed-loop systems.</li> <li class="ace-line ace-line old-record-id-PKU2dkSMhoPMEuxuUiclXKKAgmg" data-list="bullet">Optimize inference efficiency and deploy models using low-latency serving frameworks such as <strong data-start="1588" data-end="1596">vLLM</strong> and <strong data-start="1601" data-end="

技能关键字

职责方向