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

Even this site on mobile has issues with unnecessary horizontal scroll.

To be fair to the pnpm maintainers, they have been working on this well before the bun AI rewrite. The first commit in their rust experiment is in 2021.

https://github.com/pnpm/pn/commits/main/?after=e24c0b293f1f4...


Here is some more technical information on how this was trained, as well as a download link.

https://huggingface.co/thomsonreuters/Thomson-1.0-Small

(Full disclosure I’m a TR employee, although I had nothing to do with making this)


Full technical report PDF: https://huggingface.co/spaces/tri-fair-lab/publications/blob...

> In this report, we argue that frontier performance can be achieved by a wide range of institutions through Continual Learning on readily available open-weight models.

> As opposed to existing limited approaches such as small-scale fine-tuning, prompt engineering, or tool-augmentation with a frozen model, our Continual Learning approach takes advantage of the effectiveness of a modern mid- & post-training stack while introducing safeguards preserving both plasticity and stability at each training stage and seeking to make the minimal number of high-impact interventions on the parameters.

For the large model, Thomson is utilizing the fine tuning stack they describe in the article, running it on Snowdon 1.0-Large, which in turn is a fine tune of Qwen3.5 397B. Same thing for the small model, but it's a fine tune of Snowdon 1.1-Small, which is a fine tune of Qwen3.6 35B.

As for the small version's run:

> The full pipeline consumed approximately 1.63 × 10²³ FLOP over 35,207 B200 GPU-hours, showing that these results are achievable with compute and personnel budgets substantially lower than commonly thought.

That would amount to around a quarter to half a million dollars of spend on that run. 100k minimum, if they got a great deal.


I wonder where the other $39.5 million went?


Men in the middle wages


Thanks, that’s great, lots of details


So it's a qwen fine-tune?

I mean that's a reasonable thing to do, but then the press release shouldn't be written the way it is written.

They're not as detached from the rest as the industry as the writing suggests.

__

> It is obtained by repurposing the open-weight Qwen3.6-35B-A3B model and substantially improving it on a wide range of performance domains.

nice wording on the HF page tho. "Repurposing". Lmao


the press release says this explicitly


No it doesn't. There is no mention of Qwen at all.


they mention using an open weights model, which one it is doesn't really matter


Yep. Explicitly enough to be legally safe for sure.

Which is the thing with press releases. Would've been nice to not do the bare legal minimum tho


What isn’t clear in the press release? Reads fine to me


It’s on hugging face, I suppose one could do those themselves?

https://huggingface.co/thomsonreuters/Thomson-1.0-Small


I don’t think people read the article, I didn’t until I saw your comment. The article feels like clickbait tbh.


I think the unusual thing is that it was written in a week. I highly doubt that they read and understood all 1M lines. But if it works and people use it, what does that mean for software? Should we still care about the code that’s written? Should we even look? I’ve always thought so, but maybe I’m just biased.


I think we should care way more about what the validation story is of code. The obvious question does it all work? I'm happy to not look at any code if we have good ways to validate what is there. The other thing I care about is the architectural structure of the code. Given its a port I don't think that would have changed.



Ive always found this explanation helpful for why these are useful:

https://www.joshwcomeau.com/css/custom-css-reset/


Ehhhhh... what does this mean for the open source versions of all these libs? You could interpret some of the graphs as vite oss isn't getting rolldown. That would be disappointing but still okay.


> You could interpret some of the graphs as vite oss isn't getting rolldown

Vite already has rolldown support in the current version, it's just in alpha/test stage.


It seems to just be a plan at the moment although like you say you can test it out:

https://vite.dev/guide/rolldown.html#how-to-try-rolldown

Nothing is keeping them to this plan other though, I hope they do follow through. That would make the graph on the page misleading in the other direction though as the speed feature would be included in the non plus version.

I want to also say I'm a happy vite user (and the other projects that team makes).


Nothing changes on VoidZero's commitment to open source. Vite 8 is still set to get Rolldown. I mentioned that also in my talks (e.g. https://youtu.be/fnyK-xXxVKU?t=3027)


Vite OSS is getting Rolldown; it will be part of Vite from version 8 onward. Rolldown itself is also 100% OSS.


I’ve used this before, it helps when you format the entire repo and remove the one commit from the history https://docs.github.com/en/repositories/working-with-files/u...


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

Search: