A while back I reviewed the Viltrox EF-FX1 adapter, my first attempt at bridging my old Canon EF and EF-S lenses over to my Fujifilm X-S10. The abridged version of that experience: it was frustrating. Firmware roulette, random errors, camera freezes requiring battery pulls, and auto focus performance that varied wildly between versions. I kept it because it was just good enough, but it never stopped feeling like a workaround rather than a solution. Moreover, Viltrox never released any further firmware versions, so the 2.29 blob tested in that post really is the last firmware.

Fast forward to recently, and I decided to give the whole thing another shot. This time with the 7Artisans EF-FX adapter. Same concept: an electronic adapter with auto focus support, aperture control, and EXIF data transmission, and at 119€ it sits right in the same price bracket as the Viltrox. But is it actually better? Spoiler: yes. Let me walk you through it.

I like running my own LLMs locally. Open models are becoming more and more powerful, with exciting releases like the latest Qwen 3.5 family scoring highly in benchmarks even in their smaller variants. This makes managing and running your own models more viable, as it becomes increasingly easy to repurpose old hardware for local inference with progressively better results. For local users and modest purposes, the GGUF format introduced by llama.cpp is the de-facto default.

Since local inference is typically heavily restricted by the available hardware, several optimization techniques have been implemented to make the models leaner and faster. Perhaps the most important of these is quantization, which trims down the bit count per parameter to achieve lower memory usage and (sometimes) faster inference. The challenge is that there are many different formats and strategies for quantization. In this post, I summarize them, providing a bird’s-eye view on the available techniques, their strengths, and their weaknesses.

The release of LM Studio 0.4.5 has introduced a much needed feature in this local LLM suite that has it much more attractive with respect to other similar projects. LM Link allows you to connect multiple LM Studio instances across your network to share models and perform inference seamlessly.

I finished my previous post on Play Kid, only two days ago, with the following words:

Next, I’ll probably think about adding Game Boy Color support, but not before taking some time off from this project.

Yeah, this was a lie.

In my previous post, I shared the journey of building Play Kid, my Game Boy emulator. At the time, I was using SDL2 to handle the “heavy lifting” of graphics, audio, and input. This was released as v0.1.0. It worked, and it worked well, but it always felt a bit like a “guest” in the Rust ecosystem. SDL2 is a C library at heart, and while the Rust wrappers are good, they bring along some baggage like shared library dependencies and difficult integration with Rust-native UI frameworks.

So I decided to perform a heart transplant on Play Kid. For version v0.2.0 I’ve moved away from SDL2 entirely, replacing it with a stack of modern, native Rust libraries: wgpu, pixels, egui, winit, rodio, and gilrs:

When I was a kid my parents got me and my brothers a brand new Game Boy. I used to play it a lot, and I was mesmerized and amazed at how such a tiny brick could give life to so many wonderful worlds. Some of my favorite games were Super Mario Land (1, 2, and 3, Wario Land), The Amazing Spider-Man, and Kirby’s Dream Land. We also had some other games, like Bugs Bunny Crazy Castle or Tetris, that I also played quite a lot. I remember fondly too.