Daniele Salatti

@danielesalatti.com did:plc:r2skelzh7excleps23iselts

Here’s to now! LLMs, AI, Software Engineering, Mountaineering, Sailing Seattle, USA ↔ Livorno, Italy danielesalatti.it

No Bluesky desde 25 maio 2023

Seguidores: 601
Seguindo: 433
Posts: 1.232
Dias nos rankings: 0

Resumo

Daniele Salatti (@danielesalatti.com) é um usuário do Bluesky desde 25 maio 2023. Tem 601 seguidores, segue 433 contas e publicou 1.232 posts.

Aparições nos rankings

Os melhores dias em que posts desta conta entraram na lista diária do Debya.

Esta conta ainda não apareceu em um ranking diário.

Posts curtidos recentemente

Curtidas públicas da rede Bluesky.

jake@yetanotheruseless.com28/10/2026 21:46
I do not have that problem, as I split across Ant/OAI/GOOG/some long tail of OpenRouter GLM/DS models to keep making sure I know how well they do. (my problem, instead, is spending giant gob-tons on tokens!)
Ver post
mr. TIM@timkellogg.me23/10/2026 16:32
memory bandwidth is the typical constraint LLMs are so large that the they have to be paged close to the compute cores. When you batch, you can process several requests in one single matrix multiply operation so you spend less time moving weights around, more doing actual computations
Ver post
mr. TIM@timkellogg.me23/10/2026 16:32
imo speculative decoding is such a huge win for local LLMs keeping a GPU busy basically boils down to batching, but on a laptop you don’t have multiple users whereas SD allows you to batch across sequential tokens (basically) which lets you maximize the value you’re getting out of your GPU
Ver post
brennan@brennan.computer23/10/2026 16:24
I've owned and assembled a heck of a lot of it, it's a marvel of design and manufacturing and logistics as I get older their solid wood options are often nice, I'm just sort of phasing out of my cardboard era their apartment sizing schtick is now too small as I moved out of my apartment(s)
Ver post
antirez@antirez.bsky.social20/10/2026 18:50
DeepSeek infer is the cloud version of DwarfStar, conceptually. Vertical inference engine for one model and specific hardware. Good move. And well, a lesson to the ones that criticized this approach as well.
Ver post