Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

part2: Move embedding quantization kernels to fbgemm for better sharing between C2/PT #425

Closed
wants to merge 1 commit into from

Conversation

dskhudia
Copy link
Contributor

@dskhudia dskhudia commented Sep 1, 2020

Summary:
8bit with float scale and bias.

Test and benchmark added.

With scale and bias as float
bit_rate,   rows,  cols,  elems_per_usec,    GB/Sec
       8,   100,     16,          556.20,       2.22
       8,   100,     64,         1022.51,       4.09
       8,   100,    128,         1121.43,       4.49
       8,   100,    256,         1292.61,       5.17
       8,   100,    512,         1526.69,       6.11
       8,   100,   1024,         1407.09,       5.63
       8,   100,   2048,         1620.34,       6.48
       8,   120,     16,          562.60,       2.25
       8,   120,     64,         1058.52,       4.23
       8,   120,    128,         1082.74,       4.33
       8,   120,    256,         1382.87,       5.53
       8,   120,    512,         1513.15,       6.05
       8,   120,   1024,         1441.19,       5.76
       8,   120,   2048,         1634.99,       6.54
       8,  1000,     16,          598.05,       2.39
       8,  1000,     64,         1151.16,       4.60
       8,  1000,    128,         1071.58,       4.29
       8,  1000,    256,         1278.66,       5.11
       8,  1000,    512,         1441.13,       5.76
       8,  1000,   1024,         1605.48,       6.42
       8,  1000,   2048,         1764.24,       7.06

Differential Revision: D23455486

…ng between C2/PT

Summary:
8bit with float scale and bias.

Test and benchmark added.

```
With scale and bias as float
bit_rate,   rows,  cols,  elems_per_usec,    GB/Sec
       8,   100,     16,          556.20,       2.22
       8,   100,     64,         1022.51,       4.09
       8,   100,    128,         1121.43,       4.49
       8,   100,    256,         1292.61,       5.17
       8,   100,    512,         1526.69,       6.11
       8,   100,   1024,         1407.09,       5.63
       8,   100,   2048,         1620.34,       6.48
       8,   120,     16,          562.60,       2.25
       8,   120,     64,         1058.52,       4.23
       8,   120,    128,         1082.74,       4.33
       8,   120,    256,         1382.87,       5.53
       8,   120,    512,         1513.15,       6.05
       8,   120,   1024,         1441.19,       5.76
       8,   120,   2048,         1634.99,       6.54
       8,  1000,     16,          598.05,       2.39
       8,  1000,     64,         1151.16,       4.60
       8,  1000,    128,         1071.58,       4.29
       8,  1000,    256,         1278.66,       5.11
       8,  1000,    512,         1441.13,       5.76
       8,  1000,   1024,         1605.48,       6.42
       8,  1000,   2048,         1764.24,       7.06
```

Differential Revision: D23455486

fbshipit-source-id: 88ec5c220c3809d9d22230ea1dbbb55a4148a465
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D23455486

@facebook-github-bot
Copy link
Contributor

This pull request has been merged in 7e401c0.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants