Together AI Workshop: Upgrading and Customizing Open Models

​Together AI is organizing our first workshop in Amsterdam, inviting everyone excited about open LLMs! In this workshop, we will cover techniques for improving quality, speed, and costs of running open models. The Model Shaping team at Together, pushing both our research and product efforts in this field, will overview approaches for model customization and share their insights.

​Meet our Europe team, as well as fellow engineers and researchers from the Netherlands AI community, to discuss the latest advances in the AI space and learn more about Together.

​At the workshop, you will learn more about:

• ​Tailoring open models for downstream applications. By employing a range of post-training steps, such as SFT on domain-specific data, preference optimization, or training with verifiable rewards, we can obtain models that rival the strongest proprietary LLMs on target tasks at a fraction of cost.

• ​Customized speculative decoding. We can speed up inference by running a small domain-optimized draft model that generates tokens ahead of the large target model. At Together, we employ both base and custom speculators, boosting the speed of models such as DeepSeek-R1 by over 1.85x.

• ​Model quantization. Through a variety of model compression techniques, we can make even the largest LLMs require less hardware resources, reducing their inference costs. In some cases, it becomes possible to run top-ranking models on your computer or even a mobile phone!

​
Agenda:

• ​17:45–18:00 Welcome reception

• ​18:00–19:30 Talks on Open Model Customization

• ​19:50–21:00 Networking

​
⚠️ Note - Only company (non-Gmail etc.) emails will be accepted. Please provide your valid corporate email upon registration!