Hacker Newsnew | past | comments | ask | show | jobs | submit | danielhanchen's commentslogin

We made something called Divergence-300 @32 (and later @512) which tests actual inference across 32 tokens on a held out test (Terminal Bench, DeepSWE, Math etc)

We do plan to do larger benchmark suites though!


Great to hear that you are planning larger benchmarks! I am particularly interested in longer-running tasks with many steps and self-correction. Divergence is fine as long as the model can still solve the task, which Divergence-300 @32 does not measure.

The current benchmark suites that frontier AI labs use are probably a good fit, e.g.

https://z.ai/blog/glm-5.3#:~:text=Performance%20across%20com...

https://www.kimi.ai/ai-models/kimi-k3#:~:text=Performance%20...

https://www.anthropic.com/news/claude-opus-5

https://openai.com/index/gpt-5-6/

But guessing from your current benchmarks, I assume that you are severely compute-constrained. What is your time budget?


Hey we did not remove the MTP for sizes above 8GiB - but yes for small GGUFs under 8 ish GiB, we removed the MTP module (IQ2_XXS and lower), because it's 500MiB to 750MiB in size, and on small 8 GiB machines, even 500MiB is needed.

As someone in the comments said we made a separate Q4_0 MTP if that's helpful so you can use that.

But I would suggest using UD-IQ3_XXS for 10.9GB for 16GB machines or Q2_K_XL


Daniel, question I got the Qwen3.8-27B-UD-Q2_K_XL.gguf from https://huggingface.co/unsloth/Qwen3.8-27B-GGUF?show_file_in... and continue with my testing, but the model quickly felt into a loop of asking the same thing over and over again, I have seen the MOE do that but not the dense ones.

And I had similar experiences when Qwen3.8-27B unsloth images just came out with the full Q8_K_XL, I'm using an AMD setup which has modifications to save to disk the kv, but your (assuming you are part of the unsloth team) for some reason have been giving me similar issues.

I tried https://huggingface.co/mradermacher/Qwen3.8-27B-Uncensored-G... the 8 bit, 6 and 2 bit... the 2 bit almost use the complete KV doing it's thing and didn't loop itself.

It can be something in my setup, there is a very high chance of that, but the previous 3.6 images from qwen, the 27B, the 31A3 and 122 they are all unsloth and did work on my setup without issues...

Again could be my setup... let me know if there is any data I can supply to you to debug if needed.


Are you using the recommended settings for temperature and such? https://unsloth.ai/docs/models/qwen3.8#recommended-settings

Often times I run into issues like this it’s because I am using settings for a different model or just forget to set them up.


>But I would suggest using UD-IQ3_XXS for 10.9GB for 16GB machines or Q2_K_XL

For those of us with a 16GB GPU, how do they compare with ExllamaV4 at 4-bit (4.0bpw)?

It looks like that fits in 12.5GB of VRAM since embedding are left in DRAM, Unsloth Studio and other llama.cpp derivatives have to load these weights in VRAM for tied embedding models like Qwen3.8.

ExllamaV3 4.0bpw fits in 12.5G of VRAM and beats IQ4_XS according to the measurements here: [turboderp/Qwen3.8-27B-exl3](https://huggingface.co/turboderp/Qwen3.8-27B-exl3)

But those were compared against UD2.0 I guess. Also plans to support these (SOTA) quants in Unsloth Studio?


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%+)


That wasn't our problem right? Gemma officially updated tool calling which we adopted


We also made NVFP4 ones if that helps! https://huggingface.co/unsloth/Qwen3.8-27B-NVFP4


This is the version we'll be testing on our rtx 6000 today! Thank you


Why not just run FP8 on vLLM with that much vRAM? It's plenty fast.


For high concurrency, using the blackwell's native native W4A4 MLP compute path, nvfp4 is something like a 1.2-1.5x performance increase over FP8. We're doing data enrichment (so, tasks completed successfully + tokens/second) so the performance bump shows up in the tasks/month number.

I am just now getting the benchmarks running against 3.8 27b but I expect similar results from benching 3.6 27b at the same quant.


I see. Did you see any intelligence degradation between FP8 and NVFP4 for 3.6 27B? You're using vLLM, right?


I didn't have time to run this as well, but we're getting a 99% agent completion rate across all tasks and 98% task decision that matches the human selected option(s) on Qwen 3.8 27B @ NVFP4. There may be a difference between FP8 and NVFP4 but it's inconsequential for our data enrichment purposes.


We will investigate Ling!


Hey :)


Thanks haha


Hey yes - if you could describe what the issues are - we will gladly fix them!


Mea-culpa, dry-multiplier generated crap and even more so on Gemma4.


Ok no worries - if there are any future issues - feel free to message / make a HF issue - we'll fix promptly!

Also note its best to follow Gemma4's official sampling params since they evaled with it - dry multiplier sometimes works, but it actually screws up reasoning sometimes


Hey sorry what are the problems that you're experiencing - we're more than happy to help fix them!


Thank you for what you are doing.


Thanks for the support and to the community!


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: