The absolute best way to prove this works is by releasing a model that was fine-tuned with this method and then showing benchmarks depicting the improvement delta between the base model and the fine tuned one.
The work is not done. Then release it to the masses and wait a few days for the actual real world anecdotes.
Until then, this is noise.
teravor 1 minutes ago [-]
[flagged]
dang 29 seconds ago [-]
"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."
Yeah, this is just slop. No benchmarks, no concrete case studies, just some vibecoded "platform" to finetune models on your own traces.
Which is an idea that has some value, but also some weaknesses. And this implementation of it isn't forthcoming with that concept. You have to really dig in to understand what they're even talking about.
irishcoffee 3 minutes ago [-]
Benchmarks are the ultimate consolidation of halnons razor.
adrianco 14 minutes ago [-]
Local models need to be tuned to work well so this looks useful. Seems to be for general purpose model serving. I’ve been using https://github.com/adrianco/retort to run experiments for coding models across 13 different programming languages to see which frontier and local models work.
SilenN 8 minutes ago [-]
That's cool, thanks for sharing!
jack_pp 17 minutes ago [-]
Not sure I get it. The model you're improving is local? If so how do you even calculate cost compared to an API
SilenN 2 minutes ago [-]
Open source models.
wmo routes requests between frontier models and open source models that continuously train using Tinker. As the smaller models improve, more traffic gets routed to them.
Calculating cost is just tokens in/out.
digitaltrees 15 minutes ago [-]
Cool project
yiyingzhang 15 minutes ago [-]
Cool idea! How do you guarantee privacy?
SilenN 40 seconds ago [-]
It's open source!
We do have a platform we'll be launching as well to manage training + serving for you which will require more diligent privacy guarantees.
rglover 19 minutes ago [-]
Excited to play with this.
Rendered at 00:22:51 GMT+0000 (Coordinated Universal Time) with Vercel.
The work is not done. Then release it to the masses and wait a few days for the actual real world anecdotes.
Until then, this is noise.
https://news.ycombinator.com/newsguidelines.html
Which is an idea that has some value, but also some weaknesses. And this implementation of it isn't forthcoming with that concept. You have to really dig in to understand what they're even talking about.
wmo routes requests between frontier models and open source models that continuously train using Tinker. As the smaller models improve, more traffic gets routed to them.
Calculating cost is just tokens in/out.
We do have a platform we'll be launching as well to manage training + serving for you which will require more diligent privacy guarantees.