[HN Gopher] Sky-T1: Train your own O1 preview model within $450
___________________________________________________________________
Sky-T1: Train your own O1 preview model within $450
Author : fofoz
Score : 22 points
Date : 2025-01-13 08:51 UTC (14 hours ago)
(HTM) web link (novasky-ai.github.io)
(TXT) w3m dump (novasky-ai.github.io)
| elashri wrote:
| > We initially trained a 32B model using 3-4K math problems from
| the Numina dataset (provided by STILL-2), achieving a significant
| improvement in AIME24 accuracy from 16.7% to 43.3%. However, when
| we incorporated coding data generated from the APPs dataset into
| the training process, AIME24 accuracy dropped to 36.7%. We
| hypothesize that this decline is due to the distinct reasoning
| approaches required for math and coding tasks.
|
| This is interesting. For large models that were trained on much
| more data. I wonder if o1 is trained in a different way that
| GPT-4o. Do they only rely on synthetic data (plus some hand
| crafted datasets). But then how would O1 knows a lot of facts
| like GPT-4o indicating that these were in the training.
|
| Can someone with more understanding and knowledge weight on this?
| zamadatix wrote:
| Fine tune an existing model. Training such a model so cheaply
| would've been nuts.
| ipsum2 wrote:
| For math only.
| thot_experiment wrote:
| They finetuned QwQ to perform well on a benchmark. For the past
| two years there has been a constant stream of "X fine-tune beats
| Y closed model on Z benchmark". This isn't interesting and has
| never been interesting, see Goodhart's Law. If you're actually
| using local models day to day you will quickly find that
| finetunes are almost universally a waste of time. Even when it
| comes to something like smutty roleplay gaslighting a model can
| often lead to more interesting and consistent results because
| finetunes are basically always overfit to the training data.
___________________________________________________________________
(page generated 2025-01-13 23:03 UTC)