)]}'
{
  "commit": "a8d70f3552110b907c2a8e221ac5d34708938b94",
  "tree": "e9e1c4a3909ba0c3eece36379c691286ab8fecdc",
  "parents": [
    "fb7c9108d9c78d9053c2bf7bfc2f93fc8ac2eac5"
  ],
  "author": {
    "name": "Andrey Malevich",
    "email": "amalevich@fb.com",
    "time": "Wed Feb 15 01:22:09 2017 -0800"
  },
  "committer": {
    "name": "Facebook Github Bot",
    "email": "facebook-github-bot@users.noreply.github.com",
    "time": "Wed Feb 15 16:00:44 2017 -0800"
  },
  "message": "Try to improve serialization speed for SparseNN.\n\nSummary:\nCreated some simple benchmark to test model saving speed, plus few possible\noptimization on top of it.\n\nSince we don\u0027t really want to have partial LogFileDB ever, it makes sense to\ncommit the transactions only after we\u0027ve finished serialization.\n\nAs a result in my test serialization time in my dummy test drops from\n480 seconds, to:\nSerialization time: 52.5134651661\nDeserialization time: 60.5741639137\n\nOne more really scary things that I\u0027ve found:\nit looks like load_op with load_all might actually load corrupted DBs (if they\u0027ll be truncated), so we do need to fix it really badly (save all blobs we have in the DB or even better checksum).\n\nReviewed By: dzhulgakov\n\nDifferential Revision: D4558216\n\nfbshipit-source-id: 4145c07f29b9dda527a2e57842f3abd8023d71a3\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "2b3d18110d932ef01594fcece8774190812d0849",
      "old_mode": 33188,
      "old_path": "caffe2/operators/load_save_op.h",
      "new_id": "bd928064b63d5ba8665f3faa5568c17979d36815",
      "new_mode": 33188,
      "new_path": "caffe2/operators/load_save_op.h"
    }
  ]
}