search

layer6ai-labs / TabDPT

TabDPT: Scaling Tabular Foundation Models
https://arxiv.org/abs/2410.18164
11 stars 3 forks source link

Fixed dtype issue when using faiss-gpu #3

Open TJKlein opened 1 week ago

TJKlein commented 1 week ago

Numpy vector needs to be of dtype.float32 instead of dtype.float64, otherwise FAISS (faiss-gpu) will raise the following error: 

TypeError: in method 'IndexFlat_add', argument 3 of type 'float const *'
junweima commented 2 days ago

@TJKlein I will look into this and test it out for faiss-gpu. For this released version currently we are actually using faiss-cpu not faiss-gpu for all of our experiments.