Advanced Quantization Algorithm for LLMs (opens in new tab)

(github.com)

139 pointslastdong9d ago17 comments

17 comments

You can try it with this model here: https://hugston.com/models/56tps-tested-autoround-qwen35-35b... which is really well done and can run pretty fast with ctx up to 300k. Just 11.65 GB. Get the Mmproj also for vision/image processing.

khimaros7d ago

this link feels very spammy

programjames8d ago

Anyone willing to dig through the code or papers for the actual algorithm? It looks like the GitHub and papers have not been optimized for communication.

liuliu9d ago

I am actually getting interested in QAT these days, especially for LSQ+ type, but it doesn't seem like people have done that enough in open-source world at least, for 2-bit / 3-bit OPD with LSQ+ basically.

yinksta9d ago

the industry has largely moved away from QAT because the hardware required for running a quantized model are an order of magnitude less than training/QATing the fp model.

That's why things like Autoround, GPTQ, AWQ have been so popular, you don't even need enough hardware to run the original model on gpu, just cpu is enough due to the data efficiency

liuliu8d ago

Thanks. I think it is a good explanation, but also suggests a gap. QAT to me, if done right, is the only way to recover performance for extreme quantization regime. The only thing matters of course, if whether it can work. My confidence in QAT comes from the LoRA can recover most quality misses in quantization, and that is still different from QAT for extreme quantization, so it could be very wrong. I need to try it anyway.

netdur9d ago

hmm... at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy, AutoRound pushes that to ~99.4–100.n% (??) the gap is roughly 0.1–0.7 percentage points

https://github.com/intel/auto-round/blob/main/docs/gguf_alg_...

NitpickLawyer9d ago

> at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy

I call bs on that. Not even FP8 is 99.8 in every scenario. It's close, but not quite bit exact, and to say that you reach 99% with q4 is a stretch. Maybe if all you test is really old benchmark questions that are in every training set out there, but go a bit ood and you'll see your q4 crumble. Try coding in a niche language or something. Or long context math (not 1+2 from the MATH benchmark) not in aime sets, and you'll get a few percentages of accuracy loss for each quant step.

woadwarrior019d ago

> at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy

That's a tall claim. By that measure, even NVIDIA's QAD, which is AFAIK is currently SOTA for 4-bit quantization (albeit NVFP4 instead of INT4) would be worse than Q4_K_M RTN quantization. :D

https://arxiv.org/abs/2601.20088

Der_Einzige9d ago

Claims of "we preserve 99.9999%" of accuracy are made in practically every quantization paper. The whole subfield acts like it's totally fine that they are testing on datasets that these models have fully trained on.

If we were in any other subfield doing this would be considered cheating and get your paper rejected, but the quantization community really loves to spread FUD claiming that quantization doesn't harm models

Also, similar dynamic with dense vs sparse MoE models. There's a reason we keep getting dense model releases alongside the MoEs out of China.

Quantization is not free, causes significant brain damage (especially on very long contexts), and has enough academic misconduct within it that it's actively screwing up the market. Don't believe me? Go ask your local financial analyst about the markets reaction to TurboQuant and than try to square that circle with this: https://openreview.net/forum?id=tO3ASKZlok (extreme and credible allegations of academic misconduct/fraud)

kadushka9d ago

Most quant papers I've seen usually report non-trivial degradation on standard benchmarks, like 1-10% degradation (compared to FP16/BF16). Especially when using 4 bits or lower. For example, I just opened a random paper: https://arxiv.org/pdf/2410.09426 see Table 1.

p.s. dense vs MoE: both are being released because they offer different trade-offs: at the same level of quality, MoE will use less compute, but more memory.

yinksta9d ago

the Turboquant debaucle is entirely on Google doing incredible amount of academic misconduct and general incompetence.

A) the QLJ thing they added is useless and the code they released didn't even include it since it makes the results worse

B) the blogpost about turboquant was ai generated and stated that turboquant used a polar transformation when it doesn't, so for the first 2 weeks people thought turboquant involved a transformation to polar coordinates. The reason the blogpost was wrong was probably because google tried to put their useless polarquant paper on the map by talking about it repeatedly.

C) since they don't use QLJ they wholesale copied the quantization technique from the HIGGS paper without citing it (https://arxiv.org/pdf/2411.17525)

D) the whole rabitQ thing https://openreview.net/forum?id=tO3ASKZlok and google's incompetent and tonedeaf response https://openreview.net/forum?id=tO3ASKZlok&noteId=X882cbyNNM after asking rabitQ for help, then shitting on them in their paper and ignoring their emails when they objected

alfiedotwtf8d ago

At the end of the day, AI Labs aren’t going to outrun Shannon

rhdunn9d ago

My experience is that at Q5 and lower you start to see noticeable degredation in performance/quality. It's especially noticeable at Q4 where models will easily get trapped in repeating token loops. I generally use Q6.

[1] https://medium.com/@paul.ilvez/demystifying-llm-quantization...

awestroke9d ago

Is your experience with this new quantization approach from Intel? Otherwise your comment is a bit offtopic at best, misleading at worst.

bee_rider9d ago

Because the accuracy loss is pretty small in both cases, that’s still a pretty big relative improvement. I mean it’s around twice as good, right? I’m not sure how to interpret these percentage points from a usability point of view, though.

muyuu9d ago

taking x20 to x40 the time RTN took? looking at the table at the bottom

if so that's a pretty drastic trade-off

j / k navigate · click thread line to collapse

17 comments

trilogic9d ago

khimaros7d ago

this link feels very spammy

programjames8d ago

Anyone willing to dig through the code or papers for the actual algorithm? It looks like the GitHub and papers have not been optimized for communication.

liuliu9d ago

yinksta9d ago

the industry has largely moved away from QAT because the hardware required for running a quantized model are an order of magnitude less than training/QATing the fp model.

That's why things like Autoround, GPTQ, AWQ have been so popular, you don't even need enough hardware to run the original model on gpu, just cpu is enough due to the data efficiency

liuliu8d ago

netdur9d ago

hmm... at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy, AutoRound pushes that to ~99.4–100.n% (??) the gap is roughly 0.1–0.7 percentage points

https://github.com/intel/auto-round/blob/main/docs/gguf_alg_...

NitpickLawyer9d ago

> at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy

woadwarrior019d ago

> at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy

That's a tall claim. By that measure, even NVIDIA's QAD, which is AFAIK is currently SOTA for 4-bit quantization (albeit NVFP4 instead of INT4) would be worse than Q4_K_M RTN quantization. :D

https://arxiv.org/abs/2601.20088

Der_Einzige9d ago

Also, similar dynamic with dense vs sparse MoE models. There's a reason we keep getting dense model releases alongside the MoEs out of China.

kadushka9d ago

p.s. dense vs MoE: both are being released because they offer different trade-offs: at the same level of quality, MoE will use less compute, but more memory.

yinksta9d ago

the Turboquant debaucle is entirely on Google doing incredible amount of academic misconduct and general incompetence.

A) the QLJ thing they added is useless and the code they released didn't even include it since it makes the results worse

C) since they don't use QLJ they wholesale copied the quantization technique from the HIGGS paper without citing it (https://arxiv.org/pdf/2411.17525)

alfiedotwtf8d ago

At the end of the day, AI Labs aren’t going to outrun Shannon

rhdunn9d ago

[1] https://medium.com/@paul.ilvez/demystifying-llm-quantization...

awestroke9d ago

Is your experience with this new quantization approach from Intel? Otherwise your comment is a bit offtopic at best, misleading at worst.

bee_rider9d ago

muyuu9d ago

taking x20 to x40 the time RTN took? looking at the table at the bottom

if so that's a pretty drastic trade-off

j / k navigate · click thread line to collapse