Sort replies: Relevant Recent Liked
Replying to @nutlope
for just $17 we get an incredible tool saved this article
1
3
1,482
into the archive it goes
1
1,264
Replying to @nutlope
great work! love how simple it is! well done!
1
3
2,776
thank you very much!
2
2,346
Replying to @nutlope
🚀🚀
1
3
1,962
🫡🫡
1
1,741
Replying to @nutlope
This is a great article!
1
2
771
Thank you Mari!
1
679
Replying to @nutlope
nice!! need to give this a try this weekend!
1
2
1,976
would love to hear how it goes!
1
1,717
Replying to @nutlope
😻
1
1
196
🫡🙏
157
Replying to @nutlope
Can I evaluate your model on JevBench?
1
1
709
It's just a quick experimental model mostly meant as a demo so I don't have super high expectations, but yes please feel free to!
1
407
Replying to @nutlope
Great value content Hassan, as always!
1
2
4,007
Thank you Youssef!
1
3,665
Replying to @nutlope
did u do any changes on the architecture of the base model?
1
1
1,569
Nope, no changes! Just a simple finetune over Qwen3.5 4B
1,332
Replying to @nutlope
What are the inference times like? Also, does it have the same schema guarantees as Jev? Eg no hallucination on classification labels
1
2,204
Re: schema, when you use it with regex (which is the recommended way), it should always adhere to it. E2E latency will depend on how many people are hitting the model at the same time. I invite you to try it yourself: api.together.ai/models/toget…
2
1,778
Replying to @nutlope
Hassan - what benchmark did the $17 run hit?
1
1
1,473
I didn't release any benchmarks for it since it's only meant to be a demo experimental model. It was benchmarking slightly behind Jev though from what I saw. But i think the real takeaway here is you can finetune one on your data for pretty cheap & make it really good for you
2
886
Replying to @nutlope
Awesome read! Thank you.
1
1
1,205
Appreciate you reading!
1
959
Replying to @nutlope
Cool
1
814
Ty!
778
Replying to @nutlope
Fire.
1
858
Thank you brother!
810
Replying to @nutlope
Basically "local Jev" github.com/echelong/nobodywh…

GitHub - echelong/nobodywho: NobodyWho is an inference engine that lets you run LLMs locally and...

NobodyWho is an inference engine that lets you run LLMs locally and efficiently on any device. - echelong/nobodywho

github.com
187
Replying to @nutlope
This makes custom classifiers feel almost unfairly cheap
182
Replying to @nutlope
every company should build theirs
167
Replying to @nutlope
花十几刀就能训个专用小模型 以前要团队扛的分类活 现在普通人也能先跑起来 时间先省回来
234
Replying to @nutlope
love this content, will give it a try
59
Replying to @nutlope
you’re using an llm to dumb down as a classifier… makes no sense, architecture wise this sucks.
1
240
Replying to @nutlope
okay, but is it comparable in intelligence to jev?
117
Replying to @nutlope
curious how it does on a policy it's never seen. 13.5k of the examples are generated rules, so i'd expect it to be great at rules shaped like those and shakier on the messy ones a real business writes
1
200
Load more