Hi I have a statefull feature function that stores a vector in its state. When I use mert to tune the weights for my function, it doesn't proceed after three runs. I'm using a node with 256G memory for tuning, after the first run, mert uses 80% of memory, after second run, it uses 90%, and after the third run, memory usage reaches to 100%. I assume that mert runs are independent from each other, right? if so, why memory usage increases after each run? Is there any solution for this problem?
Thanks! -- Arefeh Kazemi
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
