Model
Bert Large TensorFlow2 checkpoint finetuned on Squad1.1 using seqLen=384
Use the NGC CLI to download:
Copied!
1 Version
21.02.0_ampSelected09/22/2022 5:36 PM UTC3.74 GBAccuracy: 02 EpochsBatch Size: 12GPU: A100 Copied!
21.02.0_ampSelected
09/22/2022 5:36 PM UTC3.74 GBAccuracy: 02 EpochsBatch Size: 12GPU: A100
Copied!
architecture
| Key | Value |
|---|---|
| task | QA |
| type | large |
performance
| Key | Value |
|---|---|
| training_loss | 0.619 |
| f1 score | 91.45 |
training
| Key | Value |
|---|---|
| batch_size | 12 |
| LR | 0.0004 |
| training_precision | AMP |
| dataset | SQuAD 1.1 |
| epochs | 2 |