Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -54,8 +54,12 @@ split (23,825 tokens), timing the pipe only. Warmup pass discarded.
|
|
| 54 |
|
| 55 |
| Device | Batch | Words/s |
|
| 56 |
| --- | ---: | ---: |
|
|
|
|
|
|
|
| 57 |
| cpu (Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz) | 32 | 187 |
|
| 58 |
|
|
|
|
|
|
|
| 59 |
## Sources
|
| 60 |
|
| 61 |
| Source | Author | Licence |
|
|
|
|
| 54 |
|
| 55 |
| Device | Batch | Words/s |
|
| 56 |
| --- | ---: | ---: |
|
| 57 |
+
| gpu:0 (Tesla T4, 15360 MiB) | 32 | 8,320 |
|
| 58 |
+
| cpu (Intel(R) Xeon(R) CPU @ 2.00GHz) | 32 | 336 |
|
| 59 |
| cpu (Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz) | 32 | 187 |
|
| 60 |
|
| 61 |
+
A transformer pipeline is GPU-bound: the T4 is 25x the CPU on the same machine. On CPU this runs roughly 25x slower than the `sm`/`md`/`lg` tiers, which is the price of the accuracy below.
|
| 62 |
+
|
| 63 |
## Sources
|
| 64 |
|
| 65 |
| Source | Author | Licence |
|