Hacker News Reader

Comment by scrlk | Hacker News Reader

Unsloth publishes KL divergence numbers which measures how much the quantised probability distribution changes vs unquantised: https://unsloth.ai/docs/models/qwen3.8#quantization-analysis

It's a bit bare at the moment, I assume they are going to add further detail later (eg comparison to other quants), similar to their other releases.

View on HN

The talk around KL divergence is oversold. People talk about it as if it’s not a benchmark, but at its core it is in the same neighborhood! You get a different KL divergence number on different “calibration datasets”, so in other words it’s data-dependent. It is NOT a universal guarantee about the fundamental divergence of a model.

KL divergence is nothing close to a replacement for benchmarks. As flawed as benchmarks are, KL divergence is a barely useful signal. The fact that Unsloth only just started publishing KL divergences shows how unserious the quantization space is.

Actually we do publish non KLD benchmarks - top-1% is better - for NVFP4 for eg we did MMLU Pro, GPQA, AIME 2025: https://unsloth.ai/docs/models/qwen3.6#nvfp4-benchmarks

Sometimes they're just slow and expensive, so we we KLD as a proxy measure and it's very high correlation (95%+)

> The fact that Unsloth only just started publishing KL divergences shows how unserious the quantization space is.

Just wanted to say that this is a very important point that I totally agree with. People are obsessed with KL divergence, but it is yet to be demonstrated to be a descent proxy for agentic coding benchmarks.

It's a cheap to evaluate proxy for totally broken or not, which is a good start.

It also has a lot of resolution and not a lot of noise. Better would be multi-turn benchmarks with tools but getting good precision and accuracy for that is hard and computationally expensive.

> proxy for totally broken or not

You can't really know that either.

You're very right about KL divergence. I spent a couple days playing with the Gemma 4 models. That's 10 separate models (varying weights, MoE, QAT or not, etc...) with identical tokenizers. I treated 31B at BF16 as the gold standard, feeding Wikipedia snippets, and anthropomorphizing a bit:

Gemma 4 31B: "Um, if I really said all of that, I guess I'd say this next"

Gemma 4 26B: "Dude, I would've said completely different stuff" (large divergence)

Gemma 4 12B: "Umm, there's zero chance I would've said some of this" (INFINITE divergence)

Gemma 4 E4B and E2B: "Derp derp, I'm happy to say almost anything" (lowest divergence)

For models which are chat trained, they simply would not recite Wikipedia, so the divergence is almost meaningless. I thought about capturing a realistic coding session and trying to use that as the corpus, but you need to preserve the turn-based tokens and such, so I moved on to other things.

That's not a replacement for benchmarks

Reply on news.ycombinator.com