search

lin-tan / CURE

Artefact for our ICSE 2021 paper CURE: Code-Aware Neural Machine Translation for Automatic Program Repair
Other
50 stars 17 forks source link

Train loss Nan #19

Closed HanJin996 closed 1 year ago

HanJin996 commented 1 year ago

Dear author! When I train the model with the demo data and with my own data, I got the same result: 1 epoch:1, load data:1200, lr:6.25e-05, loss:nan, apr_loss:nan, lm_loss:nan, time:6s, oom:101 val loss:nan, val apr_loss:nan, val lm_loss:nan, val ppl:nan, oom:9 ... the loss, apr_loss, and lm_loss are always nan. So, how could I solve this problem?

BEbillionaireUSD commented 3 months ago

Hi, could you please share how you solve this issue?

salted-yu commented 1 month ago

@HanJin996 Hi, could you please share how you solve this issue?

HanJin996 commented 1 month ago

Sorry, dear friends, I wasn't able to solve the problem either, and I ended up abandoning the tool.

salted-yu commented 1 month ago

Thanks for your reply.