Antonio Di Lorenzo

appassionato di tecnologia in tutte le sue forme, lavoro come sviluppatore full stack.

[Antonio Di Lorenzo @loredianto]

@devops_nk but after all i'm still here

[Antonio Di Lorenzo @loredianto]

What I see now isn't AI hype anymore; it's just AI psychosis.

[Antonio Di Lorenzo @loredianto]

@madsf88 Palantir

[Antonio Di Lorenzo @loredianto]

@ivanfioravanti Yes, if you mean software as slop.

[Antonio Di Lorenzo @loredianto]

@samongaro_ I wonder how much the job market still needs developers, even if the developer role has deeply changed.

[Antonio Di Lorenzo @loredianto]

does programming with AI still make me a developer ?

[Antonio Di Lorenzo @loredianto]

@ivanfioravanti …and, specifically, more efficient

[Antonio Di Lorenzo @loredianto]

@mayuri_3015 me trying to write code handfully today

[Antonio Di Lorenzo @loredianto]

@omkarships Sometimes I try to read it, but given the enormous amount of code produced, sometimes... I simply trust it 🥲

[Antonio Di Lorenzo @loredianto]

@NanouuSymeon ehi @grok why is importat to use terminal as a developer ?

[Antonio Di Lorenzo @loredianto]

@FaztTech sublime text with sftp plugin, nothing but git!

[Antonio Di Lorenzo @loredianto]

@antirez Super cool, Salvatore! Do you use automated development tools? If so, which ones?

[FHILY👑 @Oluwaphilemon1]

If you have an NVIDIA GPU with 16 GB VRAM, Qwen3.8-27B is about to get a lot more interesting.

And no, we’re not talking about some heavily crippled Q2 GGUF.

A new setup is coming that should let 16 GB NVIDIA cards run:

• Qwen3.8-27B
• Large context windows
• MTP speculative decoding
• Linux and Windows

That covers cards like the:

RTX 5060 Ti 16GB
RTX 5070 Ti 16GB
RTX 5080 16GB
and older NVIDIA 16GB cards.

The important part is the combination.

A 27B model + large context normally eats into VRAM pretty quickly. Then you add KV cache, multimodal components, and MTP on top.

Getting all of that working on a 16 GB card without dropping down to Q2 is the interesting bit.

For people sitting on 16 GB GPUs, this could be a pretty big upgrade.

Instead of needing 24 GB or 32 GB just to get a comfortable local setup, the 16 GB tier is about to become much more capable.

And it works on both Linux and Windows.

This is exactly the kind of local AI optimization I like seeing.

Don’t buy a bigger GPU.

Make the software use the GPU you already have.

Coming soon.

[Antonio Di Lorenzo @loredianto]

@MiaAI_lab Hi, I’m using Fedora Linux and for now I’m using Qwen3.8 27B with Q4 and KV-cache and context also quantised on my 5080. The performance is acceptable with 70k tokens for context length. I’m very interested to know what you are doing.

[Antonio Di Lorenzo @loredianto]

#NuovaFotoProfilo

[Antonio Di Lorenzo @loredianto]

a me twitter stava antipatico comunque meglio xxx

[Antonio Di Lorenzo @loredianto]

Porgi l’altro guanciale

[Elon Musk @elonmusk]

What does the future hodl?

[Antonio Di Lorenzo @loredianto]

@elonmusk pls do something for #DogeDay #dogecoin people's coin