update model card README.md

README.md

@@ -21,11 +21,11 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on the conll2003 dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.0725
-- Precision: 0.8927
-- Recall: 0.9126
-- F1: 0.9025
-- Accuracy: 0.9787
+- Loss: 0.0614
+- Precision: 0.9240
+- Recall: 0.9359
+- F1: 0.9299
+- Accuracy: 0.9837
 
 ## Model description
 
@@ -50,8 +50,8 @@ The following hyperparameters were used during training:
 - seed: 42
 - distributed_type: IPU
 - gradient_accumulation_steps: 16
-- total_train_batch_size: 64
-- total_eval_batch_size: 20
+- total_train_batch_size: 16
+- total_eval_batch_size: 5
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
 - num_epochs: 3
@@ -61,9 +61,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1     | Accuracy |
 |:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
-| 0.1168        | 1.0   | 219  | 0.0970          | 0.8545    | 0.8749 | 0.8646 | 0.9720   |
-| 0.0983        | 2.0   | 438  | 0.0739          | 0.8935    | 0.9070 | 0.9002 | 0.9781   |
-| 0.0481        | 3.0   | 657  | 0.0725          | 0.8927    | 0.9126 | 0.9025 | 0.9787   |
+| 0.0759        | 1.0   | 877  | 0.0679          | 0.9090    | 0.9181 | 0.9135 | 0.9809   |
+| 0.0577        | 2.0   | 1754 | 0.0631          | 0.9210    | 0.9316 | 0.9263 | 0.9832   |
+| 0.0176        | 3.0   | 2631 | 0.0614          | 0.9240    | 0.9359 | 0.9299 | 0.9837   |
 
 
 ### Framework versions