madduci

Are you using the right configuration for your own CPU?

On a Laptop with 32 GB RAM and Iris Xe integrated graphic card, I get between 11-18 Tokens/Second with Qwen 3.8 27B and llama.cpp with sysl Intel optimisations. Same results with the vulkan back end, although sometimes it ends in weird segmentation faults due to the memory consumption.

Interesting, what's your Context Window?

If you are using the OtelCollector, you can export logs to Grafana Loki, to have them visualised

see: https://grafana.com/docs/loki/latest/send-data/otel/

The issue was caused by Ollama. I've tried again with llama.cpp and mounting the igpu device correctly, I get 18-23 tk/s on the Iris Xe card under Debian, MUCH better than before.

good article points, but what's a good alternative Docker Image with a small footprint and a package manager to Alpine Linux, if you need to install something on it and powered by glibc?

So the money goes to families with children? Or is it going to fund other harming activities?

A Macbook Neo with an M6 and 16 GB RAM at $699/€699 would be a killer feat

Some automakers like Nissan bring their own 4G SIM, which makes the pairing of phone not important, as the head unit can access Internet by itself

The goal is that your phone will be your wallet. In EU we will have soon the EUDI Wallet, so why don't just replace your own classic wallet with phone totally?

Interesting. What are you using? I was using ollama

No, good point. I will have to tried it

Tried the 4 Bit versions. It loads, bit the <thinking> output isn't even coming out.

Really small (<100 tokens), I wanted to test its capabilities

Till now I was using successfully Qwen 3.5 and Gemma 4 at a reasonable speed

Tried yesterday on my own laptop (a UltraCore 7 255H without dedicated GPU,with 32 GB RAM), it wasn't even starting thinking, even on a small context window (65k)

I disagree. Why should we accept lower income, while billionaires are doing whatever they want, in the most illegal ways?

Unions and Fight for your rights, that's what we need

530 kWh here in July pushed to the grid with a 5kWp system

I remember a similar story shared this year at JAX2026 from the Sparkasse Group, they said they were using AI to migrate from COBOL, but they still were in the middle of the migration. Maybe they faced the same issues / problems? It seemed pretty zealous to me, that everything was working smoothly, but this article highlights the limitations

Porting it to C++ should bring also a lot of improvements in terms of.performance

I believe this is a move to make user spent even much more tokens

This happens because of capitalism.

In a few years will have a cemetery of software and products locked behind DRM or requiring a remote connection to a server that won't exist anymore, because it's too expensive to maintain. It is already happening, the effect will be much more amplified sooner

The real.answer is that people stopped making things with care and genuine interest.

Everything should be delivered in the fastest way possible, in the shortest time possible, on the tightest budget possible.

This is a recipe for creating a mess, in which AI has now pushed the limits and we are seeing how failure is now per-default embedded and shipped to users, who are now beta-testing products continuously.

This was my personal trick at the university. I was writing down all the notes during classes, even at the ugliest of my handwriting, but after classes insat down and rewrote the whole Notes in a better and clear way, so I was directly storing information in my brain.

I just needed to re-read the better notes 1 or 2 times and I could already recall the information, even at distance of time.

So what is the issue here? Distilling is still fair, on the same level like Anthropic scraped copyright protected material for their training.

So here robbers are blaming robbers?

These claims are just pointless, everytime

I run the itzg Docker Image behind a Wireguard VPN, all hosted on an embedded board (a armbian powered Radxa 12L with 16 GB Ram).

So far so good, we have an offline server, so even if the Microsoft Server is down, we can play (we use PolyMC on PC with offline accounts as well, for the same reason). By default, new users are in "spectator" mode, so they can't do anything without an activation from admin.

Friends get Wireguard QR codes, periodically we rotate them.

I wonder how much more exhausting could have been if they decided to go much southern in Italy, where the service can be much more degraded

PgBouncer is a great piece of software. We use that heavily and had made us DB operations pretty easy

Exactly, how Windows exposes the TPM functionalities is a far better experience than in other platforms

The German Healthcare System is moving to the TelematikInfrastruktur 2.0, which has foundations in Zero Trust security, which includes also remote attestation and proof with TPM.

I had made a dummy client as example and I must say that on development perspectives, it is wild.

There are some Go libraries, which support only RSA instead of EC for TPM Keys, on the C++ side you get most of results in Windows through the CryptoProvider, on Linux and Mac exists own OS solutions/access.

Implementing it isn't trivial at all, otherwise you would already seen (not vibe-coded) open source implementions spawning on GitHub.

And Compiler Explorer also