     45     _attend_fresh_diagonal (tkv/runtime/_fresh_diagonal.py:119)
      1     _worker (concurrent/futures/thread.py:89)
      1     tap_tokens_written (arbi_serve/spec_decode/dflash_tap.py:73)
      1     l2norm_fwd (fla/modules/l2norm.py:164)
      1     fn (torch/_dynamo/guards.py:2433)
