ML Researcher - Posttraining

Job description

## **About Krea** At Krea, we are building next-generation AI creative tools. We're dedicated to making AI intuitive and controllable for creatives - our mission is to build tools that empower human creativity, not replace it. We believe AI is a new medium that allows us to express ourselves through various formats - text, images, video, sound, and even 3D. We're building better, smarter, and more controllable tools to harness this medium. We recently took this a step forward with the [launch of Krea 2](https://www.krea.ai/blog/krea-2-image-model), our first foundation model, built completely from scratch for aesthetic diversity and stylistic control. We've raised over $83M and are backed by world-class investors such as a16z, Bain Capital, and Abstract. We work full-time and in-person at our waterfront office in San Francisco. We care about creativity: our team includes musicians, designers, visual artists, and engineers. We're looking for someone with deep experience in large-scale finetuning of diffusion models and posttraining techniques. We have loads of high-quality data and aim to enhance the quality and aesthetics of the models we're building. ## **Our culture** - We work full-time and in-person at our North Beach office in San Francisco. - We believe that demonstrated interest in the creative space is key: our team includes musicians, designers, visual artists and more. - Fast iteration and execution speed. Bias towards action, agency, and independence. ## **What you'll do** - Finetune diffusion models at scale to improve image aesthetics and quality. - Implement posttraining techniques ranging from supervised finetuning, preference optimization, reinforcement learning, on-policy distillation, and various distillation / acceleration techniques. - Design comprehensive eval suites and reward designs for the reinforcement learning stage focused on image space. - Train custom VLM as reward models as part of our reward design. - Train custom LLMs for prompt expansion through finetuning and reinforcement learning. - Coordinate with data teams and partners to manage collection of preference data and model evaluation results. - Work on safety alignment of our models for open source release. - Collaborate with our AI research and engineering teams to integrate advancements into our products. ## What we're looking for - Proven work of posttraining diffusion models for image or video generation. - Experience with large-scale model training, inference, and optimization. - Strong understanding of both LLM and diffusion post training pipelines and algorithms such as PPO, GRPO, DPO, OPD, and MOPD. - Strong proficiency in PyTorch and understanding of its inner workings. - Strong background in distributed training paradigms such as FSDP, CP, SP, USP, TP, and EP. Knowing how different parallelism strategies work together and their tradeoffs. - Good knowledge of low precision training / inference in FP8, NVFP4, and MXFP8. - Good understanding of algorithms and techniques used in fast inference engines such as vLLM and sglang as well as existing RL frameworks in LLM space such as slime, miles, tinker, and verl. - Understanding of various RL infrastructure and optimization techniques such as async RL, fast weight transfer, pipelining rollouts, managing off policy data. - Ability to monitor model regression and identify weak areas and turn them into concrete evals and reward design. - Experience training VLM models. Many of our custom reward models use VLM to provide reward signals for our models. - Keeping up with the developments in related fields such as LLM, VLM, representation learning, and robotics research. - Being comfortable working in a goal-oriented research environment. - Having good judgement around when one should explore different training strategies and when it's time to commit to a specific strategy to scale compute and data. - Comfortable working with underspecified goals. We expect every technical member to take an ambiguous research goal and break it down into concrete requirements, plans, experiment plan, and execution items. - Good research taste — bias towards simplicity and methods that scale well with compute, data, and minimal human supervision. ## **What we offer** - **Team**: Work alongside a world-class team building the future of AI creative tooling - **Impact**: Significant scope and company-wide impact - **Competitive compensation**: generous salary & equity packages - **Health & wellness**: 100% health & 99% dental/vision insurance premiums covered for employees, health FSA accounts, & long-term disability coverage - **Time off**: Flexible PTO policy - **Financial planning**: 401k with a 4% company-sponsored match - **Meals in the office**: breakfast, lunch, dinner - you name it, we'll cover it - **Transit**: Ubers covered to & from the office - **Sponsorship**: We're open to sponsoring international visas where we can (e.g., STEM OPT, OPT, H-1B, O-1, E-3). - And more! *Please note the above benefits & perks are for full-time employees*