mlm-4

README.md

@@ -16,7 +16,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [aubmindlab/bert-base-arabertv02-twitter](https://huggingface.co/aubmindlab/bert-base-arabertv02-twitter) on the nuha-dataset dataset.
 It achieves the following results on the evaluation set:
-- Loss: 4.2591
+- Loss: 4.3014
 
 ## Model description
 
@@ -35,26 +35,23 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 0.0001
-- train_batch_size: 32
-- eval_batch_size: 32
+- learning_rate: 1e-05
+- train_batch_size: 64
+- eval_batch_size: 64
 - seed: 42
-- gradient_accumulation_steps: 2
-- total_train_batch_size: 64
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
-- lr_scheduler_type: linear
-- lr_scheduler_warmup_steps: 2000.0
-- num_epochs: 3
+- lr_scheduler_type: constant
+- lr_scheduler_warmup_steps: 500.0
+- num_epochs: 1
 - label_smoothing_factor: 0.1
 
 ### Training results
 
 | Training Loss | Epoch | Step | Validation Loss |
 |:-------------:|:-----:|:----:|:---------------:|
-| 4.6174        | 0.64  | 500  | 4.4343          |
-| 4.3509        | 1.28  | 1000 | 4.2799          |
-| 4.3392        | 1.91  | 1500 | 4.2559          |
-| 4.2514        | 2.55  | 2000 | 4.2591          |
+| 4.6458        | 0.25  | 500  | 4.4435          |
+| 4.4927        | 0.5   | 1000 | 4.3572          |
+| 4.4381        | 0.75  | 1500 | 4.3014          |
 
 
 ### Framework versions

config.toml

@@ -1,38 +1,30 @@
 [experiment]
-name = "mlm-3"
+name = "mlm-4"
 type = "mlm"
 
 
 [dataset]
 path = "thejosango/nuha-dataset"
 dataset_revision = "main"
-with_post_text = false
 augment_ratio = 0.0
+undersampling_strategy = false
 
 
 [model]
 pretrained_model_name_or_path = "aubmindlab/bert-base-arabertv02-twitter"
 revision = "main"
-#hidden_dropout_prob = 0.2
-#attention_probs_dropout_prob = 0.2
-#classifier_dropout = 0.2
-#num_hidden_layers = 8 
-#num_attention_heads = 12
-#hidden_size = 768
-#intermediate_size= null
 
 
 [training]
-num_train_epochs = 3
-warmup_steps = 2e3
-lr_scheduler_type = "linear"
-learning_rate = 1e-4
-per_device_train_batch_size = 32 
-per_device_eval_batch_size = 32
-gradient_accumulation_steps = 2
+num_train_epochs = 1 
+warmup_steps = 5e2
+lr_scheduler_type = "constant"
+learning_rate = 1e-5
+per_device_train_batch_size = 64 
+per_device_eval_batch_size = 64
+gradient_accumulation_steps = 1 
 weight_decay = 0.01
 label_smoothing_factor = 0.1
 weighted_loss = false
-resample_data = false 
-early_stopping_patience = 0
-early_stopping_threshold = 0
+early_stopping_patience = 5
+early_stopping_threshold = 0.005

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:812c36a44c7713e58775c0aac94aa7b15c6509792d3c97038a7e4651c3cf8d29
+oid sha256:4606e079bff12241fab620c6ea27ccf1c6a4a95ad320b348353b82e869dad9ee
 size 541104501

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d94db48e3cc2b831068369c7fecfafa07c51db5aaed7a2f6a9959a46edebe263
-size 4027
+oid sha256:a6420bbe060e364db418484f88662d36095b1bb69ce021b673848354839c0617
+size 4091