been testing some "mini models" on my 3070 (8gb vram). had another to run through my benchmark today, the new bonsai 2 9b. however, this time I sat downstairs playing piano while i prompted free deepseek v4.1f via opencode through codync to complete the bench mark. it was great..
10,500 gb300s for RL post train, 8,500 of those for inference. crazy!!
Beam’s model technical report is now available. Read the full report to learn about the methodology our team used to train Beam end-to-end in-house, here: reflection.ai/documents/beam…
56
very nice, added to the reading list
2
32
todays iteration of this tweet: did the trending "redo my tv ui" via codex while I did the dishes and listened to the cameron winter live solo album. its faster and no ad slop every time I turn on the tv now. not to mention a ui that is built on actually good hci principles
been testing some "mini models" on my 3070 (8gb vram). had another to run through my benchmark today, the new bonsai 2 9b. however, this time I sat downstairs playing piano while i prompted free deepseek v4.1f via opencode through codync to complete the bench mark. it was great..
63













