Skip to content

Commit

Permalink
Update bert_guide.md
Browse files Browse the repository at this point in the history
Fix issue #276
  • Loading branch information
byshiue authored Aug 10, 2022
1 parent bb638d8 commit e1b931f
Showing 1 changed file with 9 additions and 9 deletions.
18 changes: 9 additions & 9 deletions docs/bert_guide.md
Original file line number Diff line number Diff line change
Expand Up @@ -1288,15 +1288,15 @@ User can use `export NVIDIA_TF32_OVERRIDE=1` to enforce the program run under TF
| <batch_size, seq_len> | FT Dense (ms) | FT Sparse (ms) | Dense EFF (ms) | Sparse EFF (ms) | FT Sparse Speedup | EFF Sparse Speedup |
| :-------------------: | :-----------: | :------------: | :------------: | :-------------: | :---------------: | :----------------: |
| <1, 32> | 0.80 | 0.86 | 1.30 | 1.61 | 0.84 | 0.83 |
| <1, 128> | 0.85 | 0.87 | 1.24 | 1.62 | 0.86 | 0.85 |
| <1, 384> | 1.18 | 1.18 | 1.22 | 1.68 | 0.95 | 0.86 |
| <8, 32> | 0.91 | 0.91 | 1.20 | 1.65 | 0.86 | 0.85 |
| <8, 128> | 1.66 | 1.58 | 1.59 | 1.86 | 1.05 | 0.99 |
| <8, 384> | 4.16 | 3.79 | 2.21 | 2.36 | 1.22 | 1.11 |
| <32, 32> | 1.68 | 1.58 | 1.42 | 1.77 | 1.04 | 0.91 |
| <32, 128> | 4.71 | 3.98 | 2.75 | 2.60 | 1.23 | 1.12 |
| <32, 384> | 14.79 | 12.20 | 7.90 | 7.72 | 1.17 | 1.07 |
| <1, 32> | 1.32 | 1.58 | 1.35 | 1.62 | 0.84 | 0.83 |
| <1, 128> | 1.37 | 1.59 | 1.37 | 1.61 | 0.86 | 0.85 |
| <1, 384> | 1.52 | 1.60 | 1.40 | 1.62 | 0.95 | 0.86 |
| <8, 32> | 1.39 | 1.61 | 1.39 | 1.63 | 0.86 | 0.85 |
| <8, 128> | 1.70 | 1.62 | 1.61 | 1.63 | 1.05 | 0.99 |
| <8, 384> | 3.38 | 2.78 | 2.10 | 1.89 | 1.22 | 1.11 |
| <32, 32> | 1.70 | 1.63 | 1.50 | 1.65 | 1.04 | 0.91 |
| <32, 128> | 3.62 | 2.94 | 2.28 | 2.04 | 1.23 | 1.12 |
| <32, 384> | 10.10 | 8.60 | 5.77 | 5.38 | 1.17 | 1.07 |
#### BERT performance on T4 and PyTorch
Expand Down

0 comments on commit e1b931f

Please sign in to comment.