Update README.md
Browse files
README.md
CHANGED
|
@@ -26,9 +26,9 @@ The original fp32 model comes from the fine-tuned model [Intel/bert-base-uncased
|
|
| 26 |
|
| 27 |
| |INT8|FP32|
|
| 28 |
|---|:---:|:---:|
|
| 29 |
-
| **Throughput (samples/sec)** |24.
|
| 30 |
-
| **Accuracy (eval-f1)** |0.
|
| 31 |
-
| **Model size (MB)** |
|
| 32 |
|
| 33 |
### Load with Intel® Neural Compressor (build from source):
|
| 34 |
|
|
@@ -48,7 +48,7 @@ The following hyperparameters were used during training:
|
|
| 48 |
- learning_rate: 2e-05
|
| 49 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 50 |
- lr_scheduler_type: linear
|
| 51 |
-
- num_epochs:
|
| 52 |
- train_batch_size: 8
|
| 53 |
- eval_batch_size: 8
|
| 54 |
- eval_steps: 100
|
|
|
|
| 26 |
|
| 27 |
| |INT8|FP32|
|
| 28 |
|---|:---:|:---:|
|
| 29 |
+
| **Throughput (samples/sec)** |24.922|11.202|
|
| 30 |
+
| **Accuracy (eval-f1)** |0.9142|0.9042|
|
| 31 |
+
| **Model size (MB)** |107|418|
|
| 32 |
|
| 33 |
### Load with Intel® Neural Compressor (build from source):
|
| 34 |
|
|
|
|
| 48 |
- learning_rate: 2e-05
|
| 49 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 50 |
- lr_scheduler_type: linear
|
| 51 |
+
- num_epochs: 1.0
|
| 52 |
- train_batch_size: 8
|
| 53 |
- eval_batch_size: 8
|
| 54 |
- eval_steps: 100
|