@ivanfioravanti@christopherowen Same model, same exact prompt.
https://t.co/2R1Ghg3Qkt
Avg speed: 82 tok / s
Time: 31m52s (rendered on a Mac Mini M4)
Tokens: ~9.3M, 88k out
Cache Hit: 97%
Context used: 187K
@ivanfioravanti I made an android / iOS app for deepseek harness for this reason lol. I have all my harness instances in a list, so I just select them by host and browse.
Been great for multitasking, but not great for preventing headaches from all the context switching haha.
@MiaAI_lab Very nice! Would love your feedback on a different approach I have been cooking up... https://t.co/2R1Ghg3iuV
Zero python, all rust + CUDA. Zero docker, zero kernel warmup, simple dependencies, crazy fast speeds at long context. Also 30 second model boot. Please try if you can!
We just turned nine years old. In that time we have had:
π 7 ROS 2 releases
π 11 Gazebo releases
π 71 contracts completed
π 48 Interns mentored
π 11 simulation competitions
π 5 Ignition releases
π 9 ROSCons
π 7 ROS releases
Learn more at:
https://t.co/codLztsnUJ
π¨ππ€πππ¨
Big news! ROS 2 is going to the moon! @NASA is using ROS 2 as part of the Volatiles Investigating Polar Exploration Rover (VIPER) mission. Learn all about the rover and its mission here: https://t.co/IkStnY4QLY