Model post-training, fine-tuning, reinforcement learning, preference optimization and behavioral improvements. Classified separately from open-weight model deployment and self-hosting.
Overmind, Daytona and Prima Labs host a hands-on model post-training session with agent sandboxes, training setups, lightning talks and networking.
Overmind / Daytona / Doubleword
SOMA
Participants who need to train a model must arrive at 17:00 for setup; the event begins at 17:30. Luma lists the end as 10/05 21:00, while the official calendar lists 10-05 20:00 (PDT). Check the organizer’s confirmation. Public Luma address: House of AI, 40 Boardman Pl, San Francisco, CA 94103, USA.
A discussion of agent product infrastructure, from model post-training and inference to harnesses learning from execution traces, deployment data and talent. Connects model capabilities with systems engineering.
SkyPilot and H Company discuss large-scale GPU pretraining, reinforcement learning infrastructure, concurrent agent sandboxes and cross-cloud task orchestration.
SkyPilot / H Company
SOMA
The organizer’s October 8 update says the event is full and oversubscribed; only very limited waitlist approvals may be possible depending on turnout.
Related topic · Reinforcement learning infrastructure
The reading club now covers two papers: SFT fine-tuning with sampling and Attention–FFN disaggregation for MoE serving. Roy Nissim leads the inference discussion, moderated by NVIDIA’s Shraddha Sridhar.
Leverage
Civic Center
Each paper gets a roughly 15-minute introduction followed by group discussion.
Try another date, or clear the filters to see all events on this page.
Registration details reflect the 2026-10-12 snapshot. Events past their end time are shown as ended. Check the organizer's agenda for daily hours of multi-day events.