8 ms·
Unsloth Dynamic 2.0 GGUFs
- shablulman 9mo ago[flagged]
- Maxious 9mo agoICYMI unsloth has had some major breakthroughs today with the Qwen3.5 local models https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks 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.
- jychang 9mo agoNot really breakthroughs, more like bugfixes for their broken first batch.
- danielhanchen 9mo agoNo this is false - unsure if you saw our new blog - https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks which shows SOTA on nearly all bits, and we shared all our research as well
- jychang 9mo agoYeah, I saw that yesterday. The blog post does not explain why/how the Qwen 3.5 quants uploaded on 2/27 are different from the files uploaded on 2/24. Old 2/24 Q4_K_XL commit (pre bugfix files): https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/commit/7a8e0b23fcaf1a052ad02eb73f1c0627177e8325 https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/commit/7... Questions for a postmortem that the blog post left unanswered: - Why the change? Is it just to improve PPL/KLD? Sure, we can assume PPL and KLD are not perfect benchmarks. If yes, then why change the quantization anyways? Or was the old 2/24 quant actually much worse performing in the real world?I presume the Q4_K_XL quant using mxfp4 was the issue? If the 2/24 files having a lower PPL is an actual issue due to low quality tensors, then why not just say that? - What were the main tensors that had the quantizations changed from 2/24 to 2/27? Did you now quantize attention tensors differently? Or perhaps ssm? T - What was it changed from? Was it changed from mxfp4 or q4_k to q8, or something else? A quick sentence in the blog post saying "ok, we've confirmed that using mxfp4 (or q3 or whatever) in the attention/ssm/biases/norms/etc is a bad idea, we had that in our old models on 2/24 and our new models today are better" that would make it clear. As it's written, it's trying to both say "PPL/KLD don't actually reflect real world quality" and "we changed our quant to increase PPL/KLD" at the same time, which seems contradictory.
- zargon 9mo agoExplain what about that statement is false. Your original Q4_K_XL quant was broken. People noticing that it was a total outlier among other quants is what prompted this "research". Your own data proves that your new release fixes the bugs of your original, in order to match AesSedai's PPL. Fixing bugs is great. Searching for the best quant mix is helpful. I use your quants and appreciate your work. But whitewashing this situation dilutes trust and good will.
- Kayou 9mo agoWait, 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
- segmondy 9mo agollama.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.
- pyuser583 8mo agoHow much do you use? I have lots of trouble figuring out what the limits are of a system with x amount of vram and y amounts of ram. How do you determine this?
- fc417fc802 8mo agoIdeally you'd have (parameter count) * (bits per parameter) VRAM for the entire (presumably quantized, don't forget to account for that) model. So very approximately 16 GiB for a 34B model quantized to 4 bits per parameter. You can spill to RAM in which case you at least want enough for a single active expert but really that's going to tank performance. If you're only "a bit" short of the full model the difference might not be all that large. These things are memory bandwidth limited so if you check out RAM, VRAM, and PCIe bandwidth what I wrote above should make sense. Also you should just ask your friendly local LLM these sorts of questions.
- pyuser583 8mo agoI usually do ask the llm what parameters to use. But that’s why I know so little about parameters!
- Maxious 9mo agoYep. These Mixture of Experts models are well suited for paging in only the relevant data for a certain task https://huggingface.co/blog/moe 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/ https://bknyaz.github.io/blog/2026/moe/
- mirekrusin 9mo ago2x RTX 4090, Q8, 256k context, 110 t/s
- instagib 9mo ago1 4090, Qwen3.5-35B-A3B-UD-MXFP4_MOE, 64k context, 122 t/s. Llama.cpp
- mirekrusin 8mo agoI believe it's mentioned that MXFP4 performs surprisingly bad, you may want to try other Q4s.
- danielhanchen 9mo agoOh I didn't expect this to be on HN haha - but yes for our new benchmarks for Qwen3.5, we devised a slightly different approach for quantization which we plan to roll out to all new models from now on!
- nnx 9mo agoCan you describe what is this slightly different approach and why it should work on all models?
- hedora 9mo agoNice! Your stuff ran LLMs extremely well on < $500 boxes (24-32GB ram) with iGPUS before this update. I’m eager to try it out, especially if 16GB is viable now.
- gundmc 8mo agoThe 5080 is 16GB VRAM, not system memory. I don't think you can get 24-32GB VRAM in a $500 box
- RS-232 9mo agoThat’s intriguing. I have the same card, maybe I should give it a go. Curious about your CPU/RAM/storage capacity as well. Any resources for configuring the local setup? My entire home media stack is a single compose file in a WSL distro so it would be cool if local LLM worked the same way.
- roxolotl 9mo agoWhat method are you using to do that? I’ve been playing with llama.cpp a lot lately and trying to figure out the cleanest options for getting a solid context window on 32gb vram and 64gb system ram.
- jychang 9mo ago32GB vram is more than enough for Qwen 3.5 35b You can just load the Q4_K_XL model like normal, and put all tensors on GPU without any -ot or --cpu-moe flags. If you need a massive context for some reason where model+kv cache won't fit in 32gb, then use -ot to move the ffn moe experts for 1-2 layers into RAM. You'll get a speed hit (due to loading params from slower RAM instead of fast VRAM) but it'll work.
- roxolotl 9mo agoNice ok I’ll play with that. I’m mostly just learning what’s possible. Qwen 3.5 35b has been great without any customizations but it’s interesting to learn what the options are.
- cpburns2009 9mo agoDoes llama.cpp support Qwen3.5 yet? When I tried it before, it failed saying "qwen35moe" is an unsupported architecture.
- reactordev 9mo agoYou would need the Dynamic 2.0 GGUF as discussed in the article. But mmmmmm, Q8_K_XL looks mighty nice.
- hnfong 9mo agoYes, but make sure you grab the latest llama.cpp release New model archs usually involve code changes.
- cpburns2009 9mo agoAwesome! It looks like the llama.cpp-hip AUR was updated today to b8179, and it works.
- sowbug 8mo agoIf you're running Ollama, you'll have to wait a little longer for its embedded version of llama.cpp to catch up. It can be a couple days or weeks behind.
- Havoc 9mo agoAdvances 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
- danielhanchen 9mo agoYes the new blog post https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks has some benchmarks from community people on our quants vs others on LiveCodeBench for eg!
- MarcLore 9mo ago[dead]
- aichen_dev 9mo ago[dead]
- jychang 9mo agoWhat'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 9mo agoUnsloth have just released benchmarks on how their dynamic quants perform for Qwen 3.5 https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks
- jychang 9mo agoI'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 Qwen 3.5 XL quants 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.
- danielhanchen 9mo agoDidn't expect this to be on HN haha - but sometimes HN does have older posts come up sometimes. No your conclusion is false - only the old Q4_K_XL had slightly higher perplexity, all other quants are fine. We uploaded 9TB of research artifacts to https://huggingface.co/unsloth/Qwen3.5-35B-A3B-Experiments-GGUF https://huggingface.co/unsloth/Qwen3.5-35B-A3B-Experiments-G... for the community. If you read our blog, it says KLD and PPL are actually sometimes counterintuitive - for example MiniMax some of our quants do worse on PPL and KLD vs AesSedai's one for example, but does worse on LiveCodeBench by a lot see https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks#id-3-perplexity-and-kld-can-be-misleading https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks#id-3-... This is because see https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks#id-1-some-tensors-are-very-sensitive-to-quantization https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks#id-1-... - although bitwidths are in general monotonic ie q2_k < q3_k < q4_k < q5_k etc, we find KLD and PPL are actually not monotonic ie q3_k can actually have BETTER PPL than q4_k. So the main point is bad luck on quantization - sometimes lower bits might get lower PPL and KLD, but actually this is a ruse and wrong, since on actual real world tasks, it's worse.
- electroglyph 9mo agoCheers Daniel and Mike and team, keep up the good work!
- danielhanchen 9mo agoThank you!
- tenpa0000 9mo agoI 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 9mo agoFor 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 9mo agoWhat do you use for sub-50ms inference?
- _boffin_ 8mo agoCould be bank statement line item Classification
- dyl000 9mo agoSo q6 is practically perfect, and q3 is meaningfully decent. very impressive!
- qskousen 9mo agoThis 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 https://github.com/qskousen/ggufy
- Archit3ch 9mo agoWhat's the verdict for real world use on Q3 120B (fits in 64GB) vs Q4 of a smaller model?
- FuckButtons 9mo agoBigger model wins as long as the quantization was done properly.
- deepsquirrelnet 9mo agoI love the work unsloth is doing. I only wish gguf format had better vllm support. It’s sometimes hard to find trustworthy quants that work well with vllm.
- santa_boy 9mo agoGreat timing. I downloaded the models today on LM Studio, they seem to work remarkably well. Any HN model recommendations to run on my 24GB M5 and any best practices while running them?
- raphaelmolly8 9mo ago[dead]
- roolgo 8mo ago[flagged]
- CaptainFever 8mo agoRude.
- oofbey 8mo agoWhat does it mean to say that “99.9% KL divergence” is some number like 3? In AI research and math, KL divergence is a pseudo-distance metric from one distribution to another. (Not technically a distance between two distributions because it’s asymmetric.) Folks here who spend lots of time thinking about compressing models apparently have some specific interpretation of the term. Can somebody educate me? Because I only Understand the math definition.
- oofbey 8mo agoThe confusing thing here is that there are two distributions involved here. There's the distribution over the vocabulary (possible values of each token) and the distribution over the sequence of tokens in each document. Here, the KL Divergence is calculated over the vocabulary's distribution - for a specific token, it is measuring how much the quantized model's predictions differ from the reference model. 0 means a perfect match (no loss of quality from quantizaton), and some large number like 4 nats meaning the quantized model's predictions for that token differ substantially from the reference model. The 99.9% is taken over the sequence of tokens. So it ranks all the tokens in a corpus, and it effectively finds the token with the worst predictions (relative to the reference model) out of every 1000 tokens. That's the 99.9%ile part.