More precise computation of theoretical FLOPs
Browse files
README.md
CHANGED
|
@@ -100,7 +100,7 @@ More details about the evaluation setup and the new Norwegian benchmarks will be
|
|
| 100 |
- Training precision: bfloat16
|
| 101 |
- Hardware: 256 AMD MI250X GPUs (128 GB)
|
| 102 |
- Training time: 8.5 days
|
| 103 |
-
- Theoretical computation:
|
| 104 |
- Model FLOP/s utilization (MFU): 38%
|
| 105 |
|
| 106 |
**Unique Features:**
|
|
|
|
| 100 |
- Training precision: bfloat16
|
| 101 |
- Hardware: 256 AMD MI250X GPUs (128 GB)
|
| 102 |
- Training time: 8.5 days
|
| 103 |
+
- Theoretical computation: 2.0e22 FLOP/s
|
| 104 |
- Model FLOP/s utilization (MFU): 38%
|
| 105 |
|
| 106 |
**Unique Features:**
|