AirLLM

AirLLM loads one layer at a time, so giant open models run on small GPUs (Apache-2.0). Its README lists Kimi K3 (2.8T) in 3.72 GB. The catch: it's slow.

Links

lyogavin/airllmgithub.com/lyogavin/airllm▶ Watch the Short→← All toolsWant to make Shorts like this? → AI Shorts Studio
Leverage Moves points you to free tools; all credit to their authors.
Facts checked when the Short was made.