binary-17

README.md

@@ -23,13 +23,13 @@ model-index:
     metrics:
     - name: F1
       type: f1
-      value: 0.6309942603221625
+      value: 0.6488855020003811
     - name: Precision
       type: precision
-      value: 0.5071428571428571
+      value: 0.530860349127182
     - name: Recall
       type: recall
-      value: 0.8348848603625674
+      value: 0.8343949044585988
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -39,10 +39,10 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [thejosango/nuha-mlm](https://huggingface.co/thejosango/nuha-mlm) on the nuha-dataset dataset.
 It achieves the following results on the evaluation set:
-- Loss: 1.0499
-- F1: 0.6310
-- Precision: 0.5071
-- Recall: 0.8349
+- F1: 0.6489
+- Loss: 0.6212
+- Precision: 0.5309
+- Recall: 0.8344
 - Support: None
 
 ## Model description
@@ -62,12 +62,10 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 3e-05
-- train_batch_size: 32
+- learning_rate: 5e-05
+- train_batch_size: 16
 - eval_batch_size: 32
 - 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: constant
 - lr_scheduler_warmup_steps: 1000.0
@@ -76,13 +74,28 @@ The following hyperparameters were used during training:
 
 ### Training results
 
-| Training Loss | Epoch | Step | Validation Loss | F1     | Precision | Recall | Support |
-|:-------------:|:-----:|:----:|:---------------:|:------:|:---------:|:------:|:-------:|
-| 2.4225        | 0.98  | 500  | 0.7505          | 0.5739 | 0.4681    | 0.7413 | None    |
-| 1.5207        | 1.96  | 1000 | 0.8548          | 0.5951 | 0.4935    | 0.7491 | None    |
-| 1.2637        | 2.94  | 1500 | 1.3438          | 0.5942 | 0.4475    | 0.8839 | None    |
-| 1.0813        | 3.92  | 2000 | 1.1361          | 0.6163 | 0.4792    | 0.8633 | None    |
-| 0.9742        | 4.9   | 2500 | 1.0499          | 0.6310 | 0.5071    | 0.8349 | None    |
+| Training Loss | Epoch | Step  | F1     | Validation Loss | Precision | Recall | Support |
+|:-------------:|:-----:|:-----:|:------:|:---------------:|:---------:|:------:|:-------:|
+| 3.0309        | 0.24  | 500   | 0.5091 | 0.8040          | 0.4004    | 0.6987 | None    |
+| 1.9804        | 0.49  | 1000  | 0.5306 | 0.9799          | 0.3681    | 0.9500 | None    |
+| 1.7294        | 0.73  | 1500  | 0.5238 | 1.9662          | 0.3589    | 0.9691 | None    |
+| 1.2761        | 0.98  | 2000  | 0.4732 | 0.6147          | 0.6136    | 0.3851 | None    |
+| 1.0954        | 1.22  | 2500  | 0.5614 | 0.6456          | 0.5693    | 0.5537 | None    |
+| 0.9203        | 1.47  | 3000  | 0.5684 | 0.6862          | 0.4105    | 0.9236 | None    |
+| 0.8422        | 1.71  | 3500  | 0.0    | 0.6237          | 1.0       | 0.0    | None    |
+| 0.7892        | 1.96  | 4000  | 0.3473 | 0.6111          | 0.7300    | 0.2278 | None    |
+| 0.7834        | 2.2   | 4500  | 0.4296 | 0.5738          | 0.7014    | 0.3097 | None    |
+| 0.7446        | 2.45  | 5000  | 0.5189 | 0.6224          | 0.6777    | 0.4204 | None    |
+| 0.7392        | 2.69  | 5500  | 0.6231 | 0.6188          | 0.6007    | 0.6472 | None    |
+| 0.7007        | 2.94  | 6000  | 0.6252 | 0.5837          | 0.6153    | 0.6355 | None    |
+| 0.7127        | 3.18  | 6500  | 0.6041 | 0.6996          | 0.4461    | 0.9353 | None    |
+| 0.7023        | 3.43  | 7000  | 0.5909 | 0.7383          | 0.4291    | 0.9486 | None    |
+| 0.6853        | 3.67  | 7500  | 0.5861 | 0.7998          | 0.4239    | 0.9490 | None    |
+| 0.6708        | 3.92  | 8000  | 0.6406 | 0.7828          | 0.6127    | 0.6712 | None    |
+| 0.6463        | 4.16  | 8500  | 0.5856 | 0.7472          | 0.4226    | 0.9530 | None    |
+| 0.6517        | 4.41  | 9000  | 0.6540 | 0.8006          | 0.5701    | 0.7668 | None    |
+| 0.6325        | 4.65  | 9500  | 0.6432 | 0.7416          | 0.5150    | 0.8564 | None    |
+| 0.6434        | 4.9   | 10000 | 0.6489 | 0.6212          | 0.5309    | 0.8344 | None    |
 
 
 ### Framework versions

config.toml

@@ -1,5 +1,5 @@
 [experiment]
-name = "binary-16"
+name = "binary-17"
 type = "binary"
 
 
@@ -26,10 +26,10 @@ classifier_dropout = 0.5
 num_train_epochs = 5
 warmup_steps = 1e3
 lr_scheduler_type = "constant"
-learning_rate = 3e-5
-per_device_train_batch_size = 32
+learning_rate = 5e-5
+per_device_train_batch_size = 16
 per_device_eval_batch_size = 32
-gradient_accumulation_steps = 2
+gradient_accumulation_steps = 1
 weight_decay = 0.01
 label_smoothing_factor = 0.1
 weighted_loss = false

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1912b3c6adb29b66a3968271a99b2e678dcfb3fb06bd0f14233304ddb560a271
+oid sha256:1f7ab0988aeafc4d1040d62ea16ec7830a483ebf721884a93851f20b164979a3
 size 427419953

tokenizer.json

@@ -6,7 +6,14 @@
     "strategy": "LongestFirst",
     "stride": 0
   },
-  "padding": null,
+  "padding": {
+    "strategy": "BatchLongest",
+    "direction": "Right",
+    "pad_to_multiple_of": null,
+    "pad_id": 0,
+    "pad_type_id": 0,
+    "pad_token": "[PAD]"
+  },
   "added_tokens": [
     {
       "id": 0,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc0bc99e89a204c136cec58d3fc624718e14ab6bce3ca160a06de601cc2f37c8
+oid sha256:b0bbcaae74c9fa5e9d632c695d7ac53efdc5da8d29fd14f36321d8376ec0e5ba
 size 4091