Skip to content

Instantly share code, notes, and snippets.

@persiyanov

persiyanov/w2v-bench.md Secret

Last active June 25, 2018 09:23
Show Gist options
  • Save persiyanov/c945f31ed5ff2b2244f5999be0f7b5bc to your computer and use it in GitHub Desktop.
Save persiyanov/c945f31ed5ff2b2244f5999be0f7b5bc to your computer and use it in GitHub Desktop.
Download ZIP
Original Mikolov's Word2Vec benchmark
Raw
w2v-bench.md
# workers total time processing speed (words/sec)
1 43m14.265s 73.53k
4 11m32.787s 308.00k
8 6m0.610s 670.88k
10 5m59.982s 758.9k
12 5m15.334s 862.2k
14 4m54.556s 957.74k

Parallelizes almost linearly...

@jayantj
Copy link

jayantj commented Jun 14, 2018

Interesting. One curious thing I notice is that processing_speed * total_time does not seem to be constant here.

From 1 worker to 14 workers: words/sec increases by about a factor of 13, but time taken reduces by about a factor of 9 or so. How is the processing speed measured here?

@persiyanov
Copy link
Author

@jayantj that's because vocabulary building time is also included in total time column

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment