     41     synchronize (torch/cuda/streams.py:254)
      1     _stage_rows (arbi_serve/spec_decode/dflash_driver.py:727)
      1     _replay_fingerprint (arbi_serve/cache/recurrent_pool.py:1176)
      1     release (arbi_serve/spec_decode/dflash_kv_slots.py:280)
      1     _async_verify_accept (arbi_serve/spec_decode/mtp_verify_async.py:286)
      1     append_device (arbi_serve/spec_decode/dflash_kv_slots.py:202)
