Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Hy3

hy.tencent.com

Post reply on HN
Print view View original

71–80 of 125 posts

Re: Hy3

#71

Post by ignoramous »

simonw wrote: post #33

Pelican from a few days ago: https://simonwillison.net/2026/Jul/6/hy3/ - I was using the free tier on OpenRouter, which expires on July 21st. I tried the preview model 41 days ago and got a pelican with a "change pelican color" button: https://static.simonwillison.net/static/2026/hy3-preview-pel...

Curious why TFA calls out "Tencent in China".

  tencent/Hy3. New Apache 2.0 licensed model from Tencent in China
Is there a Tencent AI lab elsewhere (MiniMax have some association with Tencent, for example)?

Re: Hy3

#72

Post by BoorishBears »

Azantys wrote: post #66
BoorishBears wrote: post #58

Earlier quoted context omitted.

Writes pretty engaging prose, finetunes well, now MIT licensed... what's not to like? Oh and very good world knowledge for the size: better than than DS4 Flash

Do people really use 100B+ models for writing? I am no writer but to me it seems like writing is one of the easiest tasks with barely any logic or reasoning and as long as its not longer than a handful of pages I expect even 8B models to perform great.

The largest model I've post-trained in the last 2 years of working on this problem was Kimi 2.5 at 1T parameters.

The simplest way I'd put it is, teaching a model to write coherently (follow rules, patterns, etc.) is easy enough: just use teacher forcing. Teaching a model to write creatively is easy enough: just use RL and punish it for not being creative.

Teaching a model to write well and creatively takes learning two partially opposing objectives that spike the learning requirements in ways that smaller models really struggle with.

Re: Hy3

#73

Post by Computer0 »

Catloafdev wrote: post #9

Curious how people feel about this compared to DS4 Flash, given they are pretty close in size. Also curious how well it holds up to heavy quantization. DS4 Flash can currently run reasonably well on systems with ~96gb+ RAM, I wonder if Hy3 can compete there.

I don’t like DS4 in my experiences with it I still prefer qwen locally and glm on api

Re: Hy3

#74

Post by fc417fc802 »

Azantys wrote: post #66
BoorishBears wrote: post #58

Earlier quoted context omitted.

Writes pretty engaging prose, finetunes well, now MIT licensed... what's not to like? Oh and very good world knowledge for the size: better than than DS4 Flash

Do people really use 100B+ models for writing? I am no writer but to me it seems like writing is one of the easiest tasks with barely any logic or reasoning and as long as its not longer than a handful of pages I expect even 8B models to perform great.

> with barely any logic or reasoning

I take it you enjoy works of literature with inconsistent world building?

Or do you mean professional as opposed to creative writing? Because the bar is even higher for that.

Re: Hy3

#75

Post by fc417fc802 »

BoorishBears wrote: post #72
Azantys wrote: post #66

Earlier quoted context omitted.

Do people really use 100B+ models for writing? I am no writer but to me it seems like writing is one of the easiest tasks with barely any logic or reasoning and as long as its not longer than a handful of pages I expect even 8B models to perform great.

The largest model I've post-trained in the last 2 years of working on this problem was Kimi 2.5 at 1T parameters. The simplest way I'd put it is, teaching a model to write coherently (follow rules, patterns, etc.) is easy enough: just use teacher forcing. Teaching a model to write creatively is easy enough: just use RL and punish it for not being creative. Teaching a model to write well and creatively takes learning…

> is easy enough: just use RL and punish it for not being creative.

How are you scoring creativity in an unsupervised manner? That seems anything but easy.

Re: Hy3

#76

Post by fc417fc802 »

docheinestages wrote: post #44
redwood wrote: post #41

Earlier quoted context omitted.

That would be great during the winter months

Why not! And free hot showers during the summer. Just looking at the progress made since 2023, I don't think the LLM architecture we have today is the most efficient. We need creative game developers to start making LLMs.

Unfortunately I don't expect indie game devs can afford high spec hardware at current prices.

Re: Hy3

#77

Post by BoorishBears »

fc417fc802 wrote: post #75
BoorishBears wrote: post #72

Earlier quoted context omitted.

The largest model I've post-trained in the last 2 years of working on this problem was Kimi 2.5 at 1T parameters. The simplest way I'd put it is, teaching a model to write coherently (follow rules, patterns, etc.) is easy enough: just use teacher forcing. Teaching a model to write creatively is easy enough: just use RL and punish it for not being creative. Teaching a model to write well and creatively takes learning…

> is easy enough: just use RL and punish it for not being creative. How are you scoring creativity in an unsupervised manner? That seems anything but easy.

Did you try reading the whole comment?

Once creativity is being measured in isolation, getting multiple responses from the model is enough to measure creativity a ton of different ways: wordfreq to identify overused phrases, getting multiple responses for the same prompt and promoting the least similar as preferred for policy optimization, etc.

But that's of limited use for stuff like getting diverse names and such. You want creativity and coherency, and if you just punish the model for using an overused phrase, the first thing it does is strongly learn a new overused phrase (or gibberish).

(Also I don't think you mean unsupervised. You probably mean without humans [since LLMs struggle to judge creativity], but that's not what unsupervised means.)

Re: Hy3

#78

Post by walrus01 »

ckocagil wrote: post #36
UncleOxidant wrote: post #16

Earlier quoted context omitted.

That's a 2-bit quant of DS4 flash. You're probably better off running Qwen3.6-27B at Q8.

Isn't Q8 way overkill these days? I see many graphs showing Q4 or Q5 having less than %1 deviation. Nvidia's NVFP4 Qwen quantization should be even better due to its better training methods.

Q8 isn't overkill if you have sufficient RAM to fit the whole model, and you care about quality. There's a number of people who have enough hardware to fit exactly one 27B to 35B size Q8 model and not more than that, so if you can fit the whole thing in Q8, no reason to use Q4 or Q6.

Re: Hy3

#79

Post by reinitctxoffset »

UncleOxidant wrote: post #16
Catloafdev wrote: post #9

Curious how people feel about this compared to DS4 Flash, given they are pretty close in size. Also curious how well it holds up to heavy quantization. DS4 Flash can currently run reasonably well on systems with ~96gb+ RAM, I wonder if Hy3 can compete there.

That's a 2-bit quant of DS4 flash. You're probably better off running Qwen3.6-27B at Q8.

I'm dusting off the local models thing after being too busy for a while and it is crazy how wrong the benchmarks and general narrative is.

First, vLLM is like, we can do better than this, we need a better default target. It parses capability wrong, silently falls back on sm_80 xmma kernels when cutlass3x_sm120_bstensorprop_... CASK kernels are available, has mad Python in the hot path, emits dubious tool call syntax, and just managed to do something so off-road it tripped a driver bug that managed to wedge MMIO so bad the EC couldn't get an SBR out so the fan was going literally max until I hard pulled power. They say AMD is worse and I'll take their word for it because patching the driver and Inductor both in the same day is plenty of grief for me.

But the amazing thing and why your comment prompted me to write all this is that Qwen3.6 is insanely good, so good I am seriously questioning the MoE dogma. I was running the NVFP4 quant with the BF16 drafter, 100+ tokens a second of clean, legible reasoning trace and flawless tool calls. It's small and you can tell it hasn't memorized half the internet, but it's reasoning is like, better than most frontier. Opus has cleaner reasoning, GPT 5.x and Gemini 3.x Pro do not. If someone scaled that boi up by 5x? I get the feeling DeepSeek did so many arch innovations in one release that they just didn't quite have the convergence, this is like, the fundamentals as artistry. It's wayyyyyy stronger than GPT-4o at over a trillion parameters.

The other thing is I was using it on OpenRouter and it was all janky in the traces, stuttering and going in circles. On another day I would have been like "what do you expect it's the size of an iPhone". I wonder how many other people have drawn that conclusion too.

I'm not going to call it a conspiracy because it's explained by neglect, but we haven't even scratched the surface of the local model ceiling. With a harness that kept the data fresh, scale up the parameter count a bit, stretch context out a bit, and write an LLM serving engine that isn't hobbyist Python jank wrapped around fuckin inductor/triton jank?

That's Claude Code Opus experience on an expensive gaming box.

Re: Hy3

#80

Post by fc417fc802 »

BoorishBears wrote: post #77
fc417fc802 wrote: post #75

Earlier quoted context omitted.

> is easy enough: just use RL and punish it for not being creative. How are you scoring creativity in an unsupervised manner? That seems anything but easy.

Did you try reading the whole comment? Once creativity is being measured in isolation, getting multiple responses from the model is enough to measure creativity a ton of different ways: wordfreq to identify overused phrases, getting multiple responses for the same prompt and promoting the least similar as preferred for policy optimization, etc. But that's of limited use for stuff like getting diverse names and such.…

I did read the the full comment and I did in fact mean exactly what I wrote when I used the term "unsupervised". I think the condescension does nothing but get in the way. Try extending the benefit of the doubt.

> enough to measure creativity a ton of different ways ...

The things you listed seem more like temperature than creativity to me. At this point it occurs to me that this is likely yet another case of highly misleading technical jargon. Suffice to say that truly creative writing requires something entirely different than unusual sentence structure - in fact it doesn't require unusual phrasing at all.

Re unsupervised, it seems the misunderstanding here follows naturally from the previous difference in word meaning. Hopefully you see the difficulty of scoring long form answers for the creativity of the underlying ideas, as well as the impossibility of using a labeled dataset to train on such a criteria.

Post reply on HN