Unsloth Dynamic 2.0 GGUFs

ICYMI unsloth has had some major breakthroughs today with the Qwen3.5 local models https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks

With the Qwen3.5 35B A3B at Q4 I've got 200k context running at 62.98 tokens per second on a local RTX5080 16GB.

▲

mirekrusin 27 minutes ago | parent | next [-]

2x RTX 4090, Q8, 256k context, 110 t/s

▲

Kayou 2 hours ago | parent | prev | next [-]

Wait, the Q4 quantization which is more than 20GB fits in your 16GB GPU ? I didn't know that was possible, I was always restricting myself to smaller model than the VRAM I had

	▲	Maxious an hour ago \| parent \| next [-]
		Yep. These Mixture of Experts models are well suited for paging in only the relevant data for a certain task https://huggingface.co/blog/moe There's some experiments of just removing or merging experts post training to shrink models even more https://bknyaz.github.io/blog/2026/moe/
	▲	segmondy 2 hours ago \| parent \| prev [-]
		llama.cpp is designed for partial offloading, the most important part of the model will be loaded into the GPU and the rest on system ram. I run 500B+ models such as DeepSeek/KimiK2.5/GLM-5 without having that much GPU vram.

▲

jychang 2 hours ago | parent | prev [-]

Not really breakthroughs, more like bugfixes for their broken first batch.

▲

qskousen 28 minutes ago | parent | prev | next [-]

This is pretty interesting, based on the blog post, it seems like they are using a technique similar to what I have been using to generate "layer sensitivity" data in my (still pretty beta) ggufy project, which is more aimed at diffusion (image) models. https://github.com/qskousen/ggufy

▲

tenpa0000 an hour ago | parent | prev | next [-]

I run Llama 3.2 3B locally for latency-sensitive classification (sub-50ms, so no room for bigger models). At that scale Q2_K vs Q4_K_M isn't just smaller — Q2 starts flipping yes/no answers that Q4 gets right. Not often, but enough to notice in production.

So the KL divergence numbers here are more useful to me than the MMLU tables honestly. I've had MMLU hold steady while the output distribution drifted enough to break things downstream.

Does the calibration dataset make much difference at 3B though? There's so little redundancy that I'd expect it to hit a floor pretty fast regardless of how good the calibration data is.

	▲	zozbot234 44 minutes ago \| parent \| next [-]
		For a simple classification task you generally want to prioritize regularization over more sophisticated behavior, so fewer parameters with larger quantization makes sense. For more generic chat-like purposes, Q2 of a larger model may often be preferable to Q4 of a smaller one.
	▲	am17an 29 minutes ago \| parent \| prev [-]
		What do you use for sub-50ms inference?

▲

Havoc 2 hours ago | parent | prev | next [-]

Advances in this space are always welcome.

I see the change in kld values is pretty modest vs prior version. Does anyone know how that translates to real world? Is more of a linear type situation or exponential etc

▲

electroglyph an hour ago | parent | prev | next [-]

Cheers Daniel and Mike and team, keep up the good work!

▲

dyl000 an hour ago | parent | prev | next [-]

So q6 is practically perfect, and q3 is meaningfully decent. very impressive!

▲

jychang an hour ago | parent | prev [-]

What's up with this post? It's a link to something which has existed for a long time, and there's a bunch of dead comments below. Some weird SEO campaign thing?

▲

tosh an hour ago | parent [-]

Unsloth have just released benchmarks on how their dynamic quants perform for Qwen 3.5

https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks

	▲	jychang an hour ago \| parent \| next [-]
		I'm aware of that, but that's not the link of the post. The post is linking to their UD 2.0 quants from a few months back. Also, the benchmarks are because they messed up the first version of their quants for Qwen 3.5 by quanting some tensors to mxfp4 that should have been in higher quality, and this is their bugfix. The post literally starts out with "We updated Qwen3.5-35B Unsloth Dynamic quants being SOTA on nearly all bits" without explaining WHY they needed to update from the original version.
	▲	lostmsu an hour ago \| parent \| prev [-]
		Looking at their benchmarks there doesn't appear to be meaningful difference between their quants and bartowsky quants.