Qwen3-235B-A22B-Thinking-2507

https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507

152•tosh•22h ago

Comments

danielhanchen•22h ago

I'm making dynamic GGUFs for local inference at https://huggingface.co/unsloth/Qwen3-235B-A22B-Thinking-2507... Also guide to run them https://docs.unsloth.ai/basics/qwen3-how-to-run-and-fine-tun...

arcanemachiner•21h ago

Would I notice a difference between the Q2_K and Q2_K_XL variants?

danielhanchen•20h ago

Oh I would always use Q2_K_XL :) It uses our dynamic methodology to quantize certain layers in different bits ie 2, 3, 4, 5, 6, 8 bits - the more important the layer is, the higher the bitrate

Squeeze2664•18h ago

How do you determine the importance of a layer in this case?

kkzz99•16h ago

Afaik they have a test bench that they use and take the activation data from that.

danielhanchen•12h ago

Yes we have around 1 to 3 million tokens of high quality self verified data that we use to calibrate models!

smallerize•15h ago

https://unsloth.ai/blog/dynamic-v2

danielhanchen•12h ago

Yes also https://unsloth.ai/blog/deepseekr1-dynamic, https://unsloth.ai/blog/dynamic-4bit, https://docs.unsloth.ai/basics/unsloth-dynamic-2.0-ggufs

DrPhish•19h ago

I generally download the safetensors and make my own GGUFs, usually at Q8_0. Is there any measurable benefit to your dynamic quants at that quant level? I looked at your dynamic quant 2.0 page, but all the charts and graphs appear to cut off at Q4.

danielhanchen•12h ago

Oh I also upload Q8_K_XL for eg, which will upcast important layers to BF16 / F16 as well!

Oh the blog at https://docs.unsloth.ai/basics/unsloth-dynamic-2.0-ggufs does talk about 1, 2, 3, 4, 5, 6 and 8bit dynamic GGUFs as well!

There definitely is a benefit for dynamically selecting layers to be at diff bit rates - I wrote about the difference between naively quantizing and selectively quantizing: https://unsloth.ai/blog/deepseekr1-dynamic

DrPhish•10h ago

Thanks Daniel. I know you upload them, but I was hoping for some solid numbers on your dynamic q8 vs a naive quant. There doesn't seem to be anything on either of those links to show improvement at those quant levels.

My gut feeling is that there's not enough benefit to outweigh the risk of putting a middleman in the chain of custody from the original model to my nvme.

However, I can't know for sure without more testing than I have the time or inclination for, which is why I was hoping there had been some analysis you could point me to.

mycpuorg•17h ago

@danielhanchen, can we use these steps to fine tune other qwen3 models too? like 480B coder or embeddings model?

danielhanchen•12h ago

Oh for finetuning - we do have some code for MoE finetuning for Qwen at https://github.com/unslothai/unsloth, but we haven't yet announced it yet!

aliljet•17h ago

I see the term 'local inference' everywhere. It's an absurd misnomer without hardware and cost defined. I can also run a coal fired power plant in my backyard, but in practice, there's no reasonable way to make that economical beyond being a toy.

(And I should add, you are a hero for doing this work, only love in my comment, but still a demand for detail$!)

regularfry•16h ago

Hardware and cost is assumed to be approximately desktop-class. If you've got a gaming rig with an RTX 4090 and 128MB RAM, you can run this if you pick the right quant.

cmpxchg8b•16h ago

128MB? Quantization has come a long way!

danielhanchen•12h ago

I think they mis-spoke 128GB* :)

regularfry•11h ago

Wishful thinking there on my part.

danielhanchen•10h ago

Though technically < 1GB is enough - one had offload it to the SSD, albeit with very slow speeds!

danielhanchen•12h ago

The trick of llama.cpp and our dynamic quants is you can actually offload the model to RAM / even an SSD! If you have GPU VRAM + RAM + SSD > the model size (say 90GB for dynamic 2bit quant), then it'll run well!

Ie you can actually run it on a local desktop or even your laptop now! You don't need a 90GB GPU for example, but say a 24GB GPU + 64GB to 128GB RAM.

The speeds are around 3 to 5 tokens / second, so still ok! I write more about improving speed for local devices here: https://docs.unsloth.ai/basics/qwen3-how-to-run-and-fine-tun...

lostmsu•15h ago

How usable are 1 and less than 1 bit quants?

danielhanchen•12h ago

Oh 1bit reminder isn't 1bit, but a mixture of 1 to 8bit! I would still use 2bit dynamic - 1bit dynamic sometimes can go into weird repetitive loops, but it still does produce reasonable output.

Larger models with 1bit do better - for eg 480B Coder 1bit actually does very well!

tosh•21h ago

If the evals hold up this is a mindblowing new weight to capability ratio

edit: afaiu deepseek r1 was 671B with 37B active params

energy123•18h ago

Am I the only one who ignores evals, unless they're holdout datasets like a new IMO competition, or at a minimum evals with a semi-private test set like ARC-AGI 2? How can we trust that these companies don't put copies of these benchmarks in their training data? They can get whatever score they want, up to 100%, easily, by just training on that data sufficiently many times.

christianqchung•18h ago

There is something of a middle ground here for benchmark skepticism. Big companies wouldn't want a massive divergence between benchmarks and real performance that people could actually notice, and I'd argue for the most part that this hasn't happened too much (although above I posted a problem with Qwen and ARC). However, finetunes by random people/groups don't carry the same downside so I'm basically skeptical of all finetunes before using them for a particular case.

energy123•18h ago

I don't believe these companies see their customers as being able to tell the difference between a real GPQA score and a GPQA score that's fudged upwards by 10%. Look at Elon Musk presenting himself to the world as a Path of Exile expert when in reality he likely hired an expert to level up his account while he himself is an amateur. They think we are idiots and will lie to us to capture market share and lock us into their ecosystem.

christianqchung•16h ago

That's true, I certainly wouldn't be able to tell. I was thinking on the order of a 20% score vs 70%, but I realize that's not a very compelling range for my point when people are boasting about <5% shifts.

nonhaver•20h ago

impressive evals. i wonder how much of that can be attributed to the enhanced context understanding. i feel like that/length are the bottleneck of the majority of commercial models.

Eisenstein•17h ago

I don't know, I think that extending context windows is actually detrimental because people assume they can just dump things in there until it fills up. You still have to deal with the limited attention that the models have, and only filling the context with things relevant to the particular thing you are trying to solve is going to be the most effective approach. If you have too much information for it to fit into a 128K window, I think you just have too much information. The entirety of Don Quixote at over 1000 pages is less than 64,000 tokens.

CamperBob2•16h ago

That sounds low by about 10x, assuming Don Quixote has 430k words (per Google).

Still, yes, I don't know of a single model that doesn't go off the rails if you actually try to take advantage of its context length specification.

Eisenstein•15h ago

Well, I loaded up Llama 3 and downloaded the novel, and for the English translation we get 545997 tokens and in the original Spanish 653981 tokens. So when I estimated it did lose a an order of magnitude. Thanks for the correction.

pama•19h ago

Anyone here has tips for the code and hardware setup to get best per-GPU throughput on H200 or B200 hardware for large reasoning traces and inputs of around 10k–40k tokens? Is there an equivalent effort to sglang’s optimization of the V3/R1 throughput for this class of models?

sophia01•19h ago

If this is actually competitive with Gemini 2.5 Pro that would be insane esp for an Apache2 truly open weights model, let's hope it's not too hacked to shine on benchmarks!

lvl155•19h ago

Qwen3 models are solid and at such a low cost, it doesn’t hurt to pair it with something like Sonnet 4 as a check. I mean it does eliminate a lot of Claude’s “You’re absolutely right!” loops.

apwell23•18h ago

> I mean it does eliminate a lot of Claude’s “You’re absolutely right!” loops.

not as scary as "Let me try a completely different approach" . Now you have to throw out all the AI slop and start from scratch.

cma•18h ago

If you aren't using source control

christianqchung•18h ago

For what it's worth, the Qwen team misreported an ARC-AGI score benchmark on the non-thinking model by a factor of 4, which has not been explained yet. They claimed a score of 41.8% on ARC-AGI 1 [0] which is much higher than what non-chain of thought models have been able to achieve (GPT 4.5 got 10%). The ARC team later benchmarked it at 11%[1], which is still a high score, but not the same as 41.8%. It's still probably a significant update on the model though.

[0] https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507

[1] https://x.com/arcprize/status/1948453132184494471

ducviet00•18h ago

Maybe 41.8% is the score of Qwen3-235B-A22B-Thinking-2507, lol. 11% for the non-thinking model is pretty high

jug•17h ago

Makes sense, it's in line with Gemini 2.5 Pro in that case. It aligns with their other results in the post.

christianqchung•16h ago

They made it very clear that they were reporting that score for the non-thinking model[0]. I still don't have any guesses as to what happened here, maybe something format related. I can't see a motivation to blatantly lie on a benchmark which would very obviously be publicly corrected.

[0] https://x.com/JustinLin610/status/1947836526853034403

mattnewton•16h ago

Could it be the public eval set vs the private eval the ARC team has? The public eval set is slightly easier and may have had a some unintentional data leakage since it was released before their training data cutoff.

coolspot•15h ago

They have provided repro for the 41.8% result here: https://github.com/QwenLM/Qwen3/tree/main/eval

Alifatisk•18h ago

Alibaba has been on fire lately, do they even sleep?

esafak•17h ago

A rhetorical question, I suppose: https://en.wikipedia.org/wiki/996_working_hour_system

Der_Einzige•17h ago

Folks who know how to train SOTA LLMs dictate their own working conditions. No one is doing 996 there unless they want to.

donedanadone•17h ago

Evals aside why are American labs not able to release open source models at the same speec?

ttul•17h ago

The Chinese labs can’t compete on inference scale because they have been prevented from accessing the most efficient chips. But since training is a mere fraction of inference these days, they can at least hurt the American companies that are generating billions via inference services.

If you can’t beat ‘em, at least pour some sand into their moat, giving China some time to perfect its own nanometer-scale fabrication. It’s a society-wide effort.

Eisenstein•17h ago

They don't release such huge open weights models because people who run open weights don't have the capability to run them effectively. Instead they concentrate on models like Gemma 3 which goes from 1B to 27B, which when quantized fits perfectly into the VRAM you can get on a consumer GPU.

regularfry•16h ago

That shouldn't be the case here. Yes, it's memory-bandwidth-limited, but this is an MOE with 22B active. As long as the whole thing fits in RAM, it should be tolerable. It's right at the limit, though.

lossolo•15h ago

> They don't release such huge open weights models because people who run open weights don't have the capability to run them effectively

This is a naive take. There are multiple firms that can host these models for you, or you can host them yourself by renting GPUs. Thousands of firms could also host open-source models independently. They don’t release them because they fear competition and losing their competitive advantage. If it weren’t for Chinese companies open-sourcing their models, we’d be limited to using closed-source, proprietary models from the U.S., especially considering the recent LLaMA fiasco.

Eisenstein•13h ago

Given the assumption that Google has Google's own interests at heart, the question isn't 'why doesn't Google release models that allow other companies to compete with them' but 'what is the reasoning behind the models they release' and that reasoning is 'for research and for people to use personally on their own hardware'.

We should be asking why Meta released the large Llama models and why the Chinese are releasing large models. I can't figure out a reason for it except prestige.

bugglebeetle•16h ago

They could, they’re just greedy, self-serving, and short-sighted. China’s doing the AI equivalent of Belt and Road to reap tremendous strategic advantages, as well as encourage large-scale domestic innovation.

adamredwoods•17h ago

Interesting, Qwen won't answer specific historical events (Tiananmen Square).

OldfieldFund•16h ago

It's made by Alibaba :)

yunohn•15h ago

Is it really that interesting to point out for every Chinese oss model release?

mceachen•15h ago

Is it not relevant to reiterate the bias (or lobotomization) for people new to this space?

lurking_swe•3h ago

No, it’s not really relevant. Should I point out that all the models from providers in the west are very “left-leaning” every time one is released? Is that helpful to the technical discussion, in any way?

If you are using an LLM for historical knowledge, questions, or research, then the chinese censorship is relevant. Or for questions about geopolitics.

ondra•12h ago

Yes. The original DeepSeek-R1 answered those questions just fine. The newer models seem to be much more brainwashed.

osti•16h ago

For the coding benchmarks, does anyone know what are OJBench and CFEval?

OldfieldFund•16h ago

Impressive evals, but... benchmarks aren't everything.

Put this prompt into qwen3-thinking, and then compare with gemini 2.5 pro:

---

As candidates for creators, we should first address chaos. What is chaos? If for a given event X in A, all possible events can occur in B, and if such independence is universal, we are faced with chaos. If, however, event X in A limits in some way what can occur in B, a relationship exists between A and B. If X in A limits B unequivocally (we flip a switch, the lamp turns on), the relationship between A and B is deterministic. If X in A limits B in such a way that after X in A, events Y or Z can occur in B, where Y occurs 40 times out of 100 after X in A, while Z occurs 60 times, then the relationship between A and B is probabilistic.

---

You have to rewrite the above acting as David Foster Wallace in 2025. Don't mention the year. Make it postmodern. Refer to current and projected events and trends. AI, robotics, etc. you have full creative control. you can make it long if you wish. change every word. make it captivating and witty. You are acting as a demiurge DFW. You need to pass the Turing test here. Sell it to the reader. Write good, high-brow fiction. Avoid phrases that are typical to LLMs/AI writers.

sophia01•13h ago

Have been using this all morning for some integral-heavy math for my PhD (trying to bound certain analytically intractable integrals). It's a bit hit-or-miss. It's been able to come up with some pretty impressive bounds but also feels like more than half it does some really dumb stuff. Compared to Gemini 2.5 Pro it's pretty solid. Its thought traces are really silly though sometimes: it'll pretend to check websites or "pull out a calculator".

Do not download the app, use the website

Open Sauce is a confoundingly brilliant Bay Area event

Turn any diagram image into an editable Draw.io file. No more redrawing

CCTV Footage Captures the First-Ever Video of an Earthquake Fault in Motion

It's time for modern CSS to kill the SPA

Show HN: Auto Favicon MCP Server

Users claim Discord's age verification can be tricked with video game characters

Simon Tatham's Portable Puzzle Collection

It's a DE9, not a DB9 (but we know what you mean)

Never write your own date parsing library

Windsurf employee #2: I was given a payout of only 1% what my shares where worth

Vanilla JavaScript support for Tailwind Plus

Efficient Computer's Electron E1 CPU – 100x more efficient than Arm?

Why MIT switched from Scheme to Python (2009)

Animated Cursors

Experimental surgery performed by AI-driven surgical robot

Why I Do Programming

What is X-Forwarded-For and when can you trust it? (2024)

Steam, Itch.io are pulling ‘porn’ games. Critics say it's a slippery slope

The future is not self-hosted

Developing our position on AI

A Union Pacific-Norfolk Southern combination would redraw the railroad map

CO2 Battery

Programming vehicles in games

Women dating safety app 'Tea' breached, users' IDs posted to 4chan

Generic Containers in C: Vec

Researchers value null results, but struggle to publish them

Steve Jobs' cabinet

Show HN: Apple Health MCP Server

Show HN: Open IT Maintenance Planner

Do not download the app, use the website

Open Sauce is a confoundingly brilliant Bay Area event

Turn any diagram image into an editable Draw.io file. No more redrawing

CCTV Footage Captures the First-Ever Video of an Earthquake Fault in Motion

It's time for modern CSS to kill the SPA

Show HN: Auto Favicon MCP Server

Users claim Discord's age verification can be tricked with video game characters

Simon Tatham's Portable Puzzle Collection

It's a DE9, not a DB9 (but we know what you mean)

Never write your own date parsing library

Windsurf employee #2: I was given a payout of only 1% what my shares where worth

Vanilla JavaScript support for Tailwind Plus

Efficient Computer's Electron E1 CPU – 100x more efficient than Arm?

Why MIT switched from Scheme to Python (2009)

Animated Cursors

Experimental surgery performed by AI-driven surgical robot

Why I Do Programming

What is X-Forwarded-For and when can you trust it? (2024)

Steam, Itch.io are pulling ‘porn’ games. Critics say it's a slippery slope

The future is not self-hosted

Developing our position on AI

A Union Pacific-Norfolk Southern combination would redraw the railroad map

CO2 Battery

Programming vehicles in games

Women dating safety app 'Tea' breached, users' IDs posted to 4chan

Generic Containers in C: Vec

Researchers value null results, but struggle to publish them

Steve Jobs' cabinet

Show HN: Apple Health MCP Server

Show HN: Open IT Maintenance Planner

Qwen3-235B-A22B-Thinking-2507

Comments