A3C model doesn't converge!

fluctuating losses. agnostic to number of parallel agents. checked loss function, everything seems fine when referenced across A3C paper and other repos. shared optimizer looks fine.
can't figure out the exact issue. 