I used the 3.6-q8 version before,
It was smarter than me, but the tasks I gave it weren't completed properly or
it would often handle things strangely.
Of course, it was still good to work with.
When 3.8 came out, I tried the 16bf(?) version first,
It seemed to get smarter, but it was too slow...
So I thought, "Well, 8-bit quantization should do it..."
So I tried 8-bit quantization, but that also felt a bit slow...
Eventually I decided, "Well, I'm not going to be coding anyway, so let me make it run faster!"
I'm running the q4 version now, and it's much better than expected.
"It finds answers by doing long inference," or so some say,
but from my perspective, even though it takes time, I can feel it does the work much better, so I'm quite happy.
Before, I asked it to generate images using ComfyUI,
It tried various things but ultimately couldn't create the image I wanted, so I just put it on hold,
But now it's running diligently, saying "Ah! I found this!"
(Of course, the results haven't come out yet, so it's hard to evaluate...)
It seems to be working hard at something, so I'm looking forward to seeing the results after work.
I'm also going to try using it with just putting qwen's head into Codex/Claude Code,
When I tried it out a bit, it worked with a different style.
Anyway, it feels like I upgraded without spending money, so I'm proud. Hehe...
