does any one try transformer models and get killed...
# neural-forecastm
marah othman
08/02/2023, 5:25 PMdoes any one try transformer models and get killed after running ?
m
Mark
08/02/2023, 5:42 PMlike tft? yes
Mark
08/02/2023, 5:42 PMwhat error are you getting, i had a heck of a time
c
Cristian (Nixtla)
08/02/2023, 5:49 PMTFT wont run in some colab examples due to limited memory on the GPUs
m
Mark
08/02/2023, 5:49 PM^ this is exactly what i ran into
Mark
08/02/2023, 5:49 PMactually it was the fact that i had 0 gpu's... added one, all good
Mark
08/02/2023, 5:50 PMjust took me a while to realize
m
marah othman
08/02/2023, 5:53 PMi am trying the models in this tutorial
https://nixtla.github.io/neuralforecast/examples/longhorizon_with_transformers.html
and i get KILLED after training a 200 epohcs
marah othman
08/02/2023, 5:56 PMI mean autoformer and informer and Patch tst
m
Mark
08/02/2023, 5:56 PMyes, what error are you getting
m
marah othman
08/02/2023, 5:56 PMKilled
marah othman
08/02/2023, 5:56 PMand stop training
marah othman
08/02/2023, 5:57 PMit should run 600 step since i put 200 step for each model but after 200 step it give me Killed and finish
m
Mark
08/02/2023, 5:58 PMseems like a hardware limitation
m
marah othman
08/02/2023, 5:59 PMhow can i know ?
m
Mark
08/02/2023, 5:59 PMtry a smaller data set, less num_samples, etc.
m
marah othman
08/02/2023, 6:00 PMnote : i am working on just cpu
m
Mark
08/02/2023, 6:00 PMyeah just on CPU did not work for me with AutoTFT
m
marah othman
08/02/2023, 6:01 PMyou mean all transformer family?
m
Mark
08/02/2023, 6:01 PMi only tried autotft
c
Cristian (Nixtla)
08/02/2023, 6:02 PM@marah othman try using GPUs. Training these transformers on CPU is painfully slow
m
marah othman
08/02/2023, 6:04 PMi have this gpu Intel Corporation CometLake-H GT2 [UHD Graphics] (rev 05) how can i make it work with my code?