It pretty much just works. Run the unsloth quant in llama.cpp and hook it up to pi. A bunch of minor annoyances like not having support for thinking effort. It also defaults to "interleaved thinking" (thinking blocks get stripped from context), set `"chat_template_kwargs": {"preserve_thinking": True},` if you interrupt the model often and don't want it to forget what it was thinking.
One interesting thing about Qwen3 is that looking at the benchmarks, the 35B-A3B models seem to be only a bit worse than the dense 27B ones. This is very different from Gemma 4, where the 26B-A4B model is much worse on several benchmarks (e.g. Codeforces, HLE) than 31B.
> This is very different from Gemma 4, where the 26B-A4B model is much worse on several benchmarks (e.g. Codeforces, HLE) than 31B.
Wouldn't you totally expect that, since 26A4B is lower on both total and active params? The more sensible comparison would pit Qwen 27B against Gemma 31B and Gemma 26A4B against Qwen 35A3B.
> I suspect a possible future of local models is extreme specialisation - you load a Python-expert model for Python coding, do your shopping with a model focused just on this task, have a model specialised in speech-to-text plus automation to run your smart home, and so on.
I'd find this very surprising, since a lot of cognitive skills are general. At least on the scale of "being trained on a lot of non-Python code improves a model's capabilities in Python", but maybe even "being trained on a lot of unrelated tasks that require perseverance improves a model's capabilities in agentic coding".
For this reason there are currently very few specialist models - training on specialized datasets just doesn't work all that well. For example, there are the tiny Jetbrains Mellum models meant for in-editor autocomplete, but even those are AFAIK merely fine-tuned on specific languages, while their pretraining dataset is mixed-language.
I'd be surprised if that was required. OpenAI's o3 was already professional-human-level at guessing location from a photo, so unless the companies intentionally stopped training on those datasets, modern models should be too.
Gore in shooters is culturally treated as much less "violent" than e.g. graphic scenes of suicide. You could make an argument that it shouldn't be, but it is.
Which is really sad. Video games with dehumanized violence: a-ok. Video games that show you the actual consequences of violence, well that is a bridge too far.
Perhaps soiety would be better if it was reversed.
Woah, that's quite an issue. The equivalent code in Python doesn't typecheck, since `list` is invariant and hence list[str] doesn't fit list[str|int]. Unusual for TS to handle types worse than Python.
This gives me an idea for an extension similar to this mod but for Firefox, for those who are insane enough to try it: 1/10000 Chance for Withered Foxy Jumpscare Every Second
For some reason that metal pipe sound was a meme a few years ago, a picture of a pipe and that sound has 5 million views on YouTube: https://www.youtube.com/watch?v=iDLmYZ5HqgM
There's lots of valid critiques of HPMOR (I recently reread it and the early chapters are painfully obnoxious), but I think "no meaningful plot" and "doesn't do anything interesting" are objectively false. It has like a dozen interacting plotlines, and is massively different from any other HP fanfic and most media in general. It is popular for a reason. If you dropped it early, I'd encourage you to try again.
reply