Joe Fioti
@joefioti
Followers
1K
Following
571
Media
99
Statuses
928
@luminal_ai (yc s25). building a compiler to make models go really fast.
San Francisco, CA
Joined June 2015
im a schizo always talking about egraphs and saturation, matt made a great explainer vid for those of you still sane.
going through our tech stack here at luminal from a high level.first one is on e-graphs
1
0
5
RT @michwirantono: Hundreds of founders move to SF every 3 months for @ycombinator, but most of us barely see the city as we’re heads down….
0
5
0
there are those who dont understand how much software matters, and there are those who will own the future.
Using May 2025 software, GB200 NVL72 had 9% worse DeepSeek 670B MoE training perf per TCO than H100. Using July 2025 software, GB200 NVL72 massively improved and now has an 1.52x better perf per TCO than h100. Performance projections suggests that in December 2025 software, GB200
0
1
14
RT @ALEXEIMARTOV: @EricJhonsa Most the time I make a request for programming it’s 50-100k tokens. It’s expensive and kinda slow. I would sp….
0
3
0
Not a joke btw.
I will pay you $10,000 if you refer me to our first hire at Luminal.
0
0
6
taking a break from kernels to write a visualizer /debugger. very cathartic
1
0
21
RT @NousResearch: Special thanks to our launch day partners - @chutes_ai, @nebiusai, and @luminal_ai - for serving these models and powerin….
0
6
0
We're live hosting the latest and greatest reasoning model from Nous!.
Nous Research presents Hermes 4, our latest line of hybrid reasoning models. Hermes 4 builds on our legacy of user-aligned models with expanded test-time compute capabilities. Special attention was given to making the models creative and interesting to
4
2
35
Still feels like egglog is this crazy dark art and we’re one of the only ones actually using it. If you’re here and know your egglog, reach out!.
0
1
15
Amazing job @stake_jevens @matthewjgunton and @dasha_shunina from @puzzlefin ! Couldn't have asked for better co-hosts. Stay tuned for more events coming soon!.
0
0
5
