News for hackers. Stuff that matters.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by robertkarl on Wednesday November 18, 2026 @03:28PM (#50551690)
I look forward to testing this against my private eval set to determine if it's benchmaxxing or actual improvement based on just 1200 tasks. downloading it now
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by xlayn on Wednesday November 18, 2026 @03:34PM (#50551775)
Snap, no gguf to download, will wait for one to appear to test, any improvement over an already good model is good news.

"18 days on 144 H200 GPUs" insane.. didn't thought so much hardware would be needed to train Qwen.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by drdaeman on Wednesday November 18, 2026 @04:39PM (#50552484)
[delayed]
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by smallerize on Wednesday November 18, 2026 @04:54PM (#50552729)
It is already quantized. You need to convert to GGUF format though. Did your assistant explain this wrong?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by drdaeman on Wednesday November 18, 2026 @04:58PM (#50552779)
[delayed]
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by cookiengineer on Wednesday November 18, 2026 @08:10PM (#50554743)
It isn't if you train the quantized model with a quantized K/V and cache. That's my approach and usually takes around a day for 100k reasoning traces (including abliteration).

Also this is on consumer hardware, meaning 96GB CPU RAM and 48GB GPU VRAM, using an optimized Vulkan concurrent shader pipeline. So costs for the machine are around 3-5k depending on RAM crisis.

[1] https://github.com/cookiengineer/qwen-reasoning-trainer

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by XCSme on Wednesday November 18, 2026 @03:35PM (#50551777)
Is it post-training or bench-maxxing?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by octoberfranklin on Wednesday November 18, 2026 @03:43PM (#50551867)
FWIW that is about $331,000 on low-onboarding-grief clouds (e.g. vast.ai) right now.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by lumost on Wednesday November 18, 2026 @04:25PM (#50552317)
Savings would likely be larger with alternative accelerators such as trainium, no?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by octoberfranklin on Wednesday November 18, 2026 @04:38PM (#50552473)
Savings would likely be larger with magic beans such as floobarbaz, no?

But seriously. If there were a cheaper and reliable way, I'm sure the authors would have mentioned it instead of H200 SXM modules. They're trying to brag about cheap training.

Maybe you wanted to use an opportunity to shill tranium.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by phs318u on Wednesday November 18, 2026 @05:02PM (#50552828)
> Maybe you wanted to use an opportunity to shill tranium.

This was unnecessary and against HN guidelines to engage in good faith.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by octoberfranklin on Wednesday November 18, 2026 @06:55PM (#50554059)
No, and we get way too many low-effort "how does this compare to $FOO" comments that serve only to improve brand recognition of $FOO.

Once you realize people are doing it you see it everywhere.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by mips_avatar on Wednesday November 18, 2026 @06:02PM (#50553604)
You'd probably end up spending more on pcie h200s. They're not that much cheaper than 8x sxm nodes and they only have 4 way nvlink. At the scale these guys were training at you'd probably want at a minimum an 8x nvlinked node for the trainer, even though you could technically fit the full training and optimizer states on a 4x h200 node.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by woah on Wednesday November 18, 2026 @04:12PM (#50552188)
The heavy reliance on eval datasets showcased here and also evidenced by the frontier lab's high payments for such data cast doubt on the claims that AI will become self-improving soon
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by charcircuit on Wednesday November 18, 2026 @05:25PM (#50553172)
>doubt on the claims that AI will become self-improving soon

It already happened. https://github.com/karpathy/autoresearch

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by its_down_again on Wednesday November 18, 2026 @05:53PM (#50553515)
Curious, where did you see the numbers on how much frontier labs are paying for eval data?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by dimes on Wednesday November 18, 2026 @06:15PM (#50553724)
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by trescenzi on Wednesday November 18, 2026 @08:16PM (#50554790)
I’ve gotten recruiting emails to be a human grader of Claude coded projects. 6 month contract at $150/hour 40hrs/week. Not insanely expansive but also not cheap.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by anon373839 on Wednesday November 18, 2026 @06:27PM (#50553820)
The frontier labs want you to believe that their models are magically gaining intelligence that generalizes to new areas with each new release. But it’s an illusion: the models are post-trained on a HUGE array of hand-curated datasets for specific tasks, with good enough coverage that it feels general.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by harry8 on Wednesday November 18, 2026 @06:53PM (#50554042)
The frontier labs are using all their customers' logs to train. It's the best training set there is.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by boardendgrain on Wednesday November 18, 2026 @07:56PM (#50554635)
Time to do all business in cockney ryming slang!
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by XenophileJKO on Wednesday November 18, 2026 @09:52PM (#50555462)
I don't think your data is as valuable as you think it is.

I would not want a model trained on user conversations.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by Bnjoroge on Wednesday November 18, 2026 @11:37PM (#50556138)
it's pretty valuable, and there's a huge demand for it from data labs depending on the kind of data. Agent traces are also pretty valuable
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by visarga on Wednesday November 18, 2026 @11:48PM (#50556183)
It is longitudinal data with LLM responses to human tasks and outcomes showing up later in the stream, you can use hindsight to see how a LLM response fared, humans come back and bring those outcomes to continue iterating.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jhancock on Wednesday November 18, 2026 @11:57PM (#50556230)
Both OpenAI and Anthropic have setting that allow (at least paid customers) to ensure their data is not used for training. Do you have reason to believe this setting is not what most user believe it is?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by harry8 on Thursday November 19, 2026 @12:22AM (#50556365)
Seriously who the f@&k actually believes that? The mind boggles that people actually say “there’s a setting.” Is it next to the one that says they won’t break the law and infringe copyright to train their models?

Sam Altman, would you actually believe anything he said if he had an incentive to lie to you?

Rieman hypothesis story was not secret. How do you even entertain this stuff for a second? Even if it were in your financial interest to not understand it’s just so very stark.

The AI companies will tell whatever lie helps them and won’t blink when caught and there will be no consequences because we’ve seen it. Repeatedly. There isn’t a grey area…

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by SubiculumCode on Thursday November 19, 2026 @12:37AM (#50556435)
Dude.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by outside1234 on Thursday November 19, 2026 @01:03AM (#50556569)
He’s also not necessarily wrong. The guy is a rat and everyone here knows it.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by ipaddr on Thursday November 19, 2026 @12:43AM (#50556463)
Both tell you not to include password/tokens in chat. Why would it matter if they weren't training
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by anon373839 on Thursday November 19, 2026 @01:02AM (#50556560)
I think if you read the legal agreements adversarially, there are a lot of cracks in them the labs can exploit. For example, what about derivative data such as LLM-generated summaries of your sessions? Do the training settings preclude use of that?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by endgame on Thursday November 19, 2026 @01:07AM (#50556596)
Anthropic used to charge you API rates if it saw a HERMES.md in the repo you were working on, even if you had a subscription: https://github.com/anthropics/claude-code/issues/53262

Another commenter here: https://news.ycombinator.com/item?id=49502774

> I made it to a third round interview at anthropic in 2024 and they had me build a web crawler as their programming test. Part way through I started on making it respect robots.txt and I could immediately tell they were no longer interested in me.

Everything but the core models seems to be some mixture of slapdash, petty, and reckless. I would absolutely believe they'd train on customer conversations. Where else would those customers go?

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jpardoe on Thursday November 19, 2026 @12:10AM (#50556291)
That depends on the customer. How can you detect the ones that are making smart decisions and guiding the agent correctly towards better outcomes? If you could know that then the problem would already be solved.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by 0xWTF on Wednesday November 18, 2026 @06:53PM (#50554044)
How is that different from you?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by StilesCrisis on Wednesday November 18, 2026 @07:16PM (#50554255)
Humans generalize much more quickly than models? We don't require millions of curated examples to learn a concept?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by p-e-w on Wednesday November 18, 2026 @08:26PM (#50554907)
Because there is four billion years of evolutionary pre-training.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by bavell on Wednesday November 18, 2026 @08:43PM (#50555030)
Instead of pre-training, I'd say it's more like a better architecture, training regime, higher fidelity data, and perhaps better initial "weights".
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by LoganDark on Wednesday November 18, 2026 @08:49PM (#50555082)
All of that stuff was still trained by evolution, though.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by gegtik on Wednesday November 18, 2026 @09:59PM (#50555513)
Hmm perhaps I learned to drive in the womb from my moms epigenetic driving genes
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by staplers on Wednesday November 18, 2026 @11:27PM (#50556081)
Monkeys couldnt be taught how to do it. Thus the analogy stands. Our brain architecture allows for more precise instructional retention
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by octoberfranklin on Wednesday November 18, 2026 @10:14PM (#50555600)
In ~8G weights of DNA?

Lol.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jodrellblank on Wednesday November 18, 2026 @10:26PM (#50555697)
Maybe not just DNA. "Something needs to speed evolution up" - Denis Noble, Professor of Biology argues against the idea that DNA is instructions and everything else in the body passively takes orders and has no input and no feedback - https://www.youtube.com/watch?v=NAPhBt8VJCM
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:28PM (#50552360)
I always wonder what you lose when you start maxing out some specific benchmark. It is really hard to do this in a balanced way and I would love if everybody that did these kind of efforts would take the time to also list the before/afters of a number of other benchmarks that they did not use in their training.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by gardnr on Wednesday November 18, 2026 @04:42PM (#50552521)
Check the article. It’s the first graphic that has other benchmarks.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:52PM (#50552697)
Yes, they all go up. I am wondering what was lost.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by nblintao on Wednesday November 18, 2026 @04:46PM (#50552603)
Good point! First of all, none of the benchmark tasks were used in training, though the training data was designed to improve capabilities on tasks like TB4.1. If you consider TB4.1 the main target, the blog actually shows before/after results on several other benchmarks, with no regressions and even some improvements.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:51PM (#50552684)
Yes, I've seen that, but I wonder what is missing because I find it hard to believe that you can just do some RL and then end up with a model that is better across the board. I would find it far more believable if you outperformed on the benchmark that was your goal (assuming you selected training data to give you an edge there) and underperformed on things that you were not optimizing for. But since it all moves in one direction (better) I feel there must be some kind of unknown at play here. Even so, task specific AI is definitely the future, you don't hire a plumber to do your bookkeeping either...
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by reissbaker on Wednesday November 18, 2026 @04:58PM (#50552785)
Why do you find it hard to believe that you can do some RL and improve across the board (albeit less strongly on other semi-related tasks than your objective)? That's the idea behind generalization: it's not perfect but you get a little bit of cross-task generalization, so if you just keep spamming more RL environments you can hill-climb to AGI. That's what the labs are doing!
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @05:05PM (#50552875)
You can hill climb but you will (or at least, I think you will!) lose generality.

And no, that won't get you to AGI, that is a completely different goal.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by banghua13 on Wednesday November 18, 2026 @05:25PM (#50553176)
If you carefully curate the data, plus the original model isn't saturated for RL post-training, it's possible to improve targeted benchmarks without regression on other metrics. This experiment is done with terminal bench 4.1 as a target, and the regression on other benchmarks has been carefully controlled through data + infra design.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @05:34PM (#50553282)
Ok, so the obvious question then is: this seems (I'm not saying it is, it just looks like a small thing compared to training up a model from scratch) a relatively trivial undertaking for the authors of the original model, if there was this much left in the can why didn't they do something similar? Or is there some step here that only you can perform? Or is it simply a matter of economics?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by mik1998 on Wednesday November 18, 2026 @06:28PM (#50553826)
[delayed]
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @07:01PM (#50554114)
That mirrors my experience exactly. I keep a couple of older models (and much larger ones as well) for when I want to do stuff that is unrelated to computer programming because the models that are really good at that tend to fare less well at other stuff. I suspect that there is a boundary where maximizing benchmarks for coding at the expense of general knowledge is going to give you a model that is really fast and good but 'narrow' and that this may also lead to loss of performance when applied to coding tasks that are less well represented in the training set (such as, for instance, DSP code).
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by hrncn on Wednesday November 18, 2026 @04:49PM (#50552656)
At least so far I am not aware of any generally usable small models that were actually improved except for Thomson Reuters qwen, but even that improvement only applies to specific topics. Everything else was pretty much a waste of time due to benchmaxxing and all kinds of weird issues caused by too much RL.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by pkoiralap on Wednesday November 18, 2026 @05:40PM (#50553366)
Sabi-noema also looks like a close competitor in the same space. Built from qwen3.8 and made better.

https://huggingface.co/sabicap/sabi-noema

Also HF model here: https://huggingface.co/RadixArk/DaisyRose-27B

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by mips_avatar on Wednesday November 18, 2026 @06:06PM (#50553651)
How did you guys set up the RL topology? Were you guys doing individual RL nodes (4xh200 rollout/4xh200 train) or did you have dedicated trainer and rollout nodes?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by nblintao on Wednesday November 18, 2026 @08:29PM (#50554933)
Dedicated rollout and trainer nodes, with RDMA transfers between them.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by 0x69420 on Wednesday November 18, 2026 @08:20PM (#50554845)
3.8 27B is an interesting choice. on the face of it, doing RL on it feels like refrying twice-cooked pork then stuffing it in the microwave for good measure. the fact that even one benchmark (tb4.1) increased more than marginally is kind of miraculous. would be interesting to see where "miles" would take eg. 3.8 flash next, which is already impressive despite being an under-trained architectural preview.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by chriddyp on Thursday November 19, 2026 @12:32AM (#50556404)
Very excited to see more post-training happening on these open models. Looking forward to trying this one out once it makes its way onto OpenRouter.

Qwen 27B really punches above its weight. 27B is small enough to fit on a Mac Mini / Macbook Pro which is very exciting.

Starting to see more promising examples of companies post-training. From my evals [1], I've tested 3 so far:

1. Fireworks' Ember-1 (postrained on Kimi K3, improved score by 10% in my evals)

2. Particle AI's Particle-1 (fine-tuned on DeepSeek V4.1 Flash, also a 10% improvement and a bit cheaper and faster)

3. Nex AGI's Nex-N2.6 (post-trained on GLM 5.3 Flash. Didn't show an improvement in my eval, and had a couple runaway loops that made it expensive).

With Thinking Machines' western open model release (Inkling Small) a few weeks ago, I expect we'll see more of this, especially out of western companies.

[1] https://plotly.com/blog/nex-n2-6-plotly-data-analytics-bench...