binary-19

README.md

@@ -23,13 +23,13 @@ model-index:
     metrics:
     - name: F1
       type: f1
-      value: 0.5023543990086741
+      value: 0.6530426884650318
     - name: Precision
       type: precision
-      value: 0.33620832642229226
+      value: 0.6085484553533643
     - name: Recall
       type: recall
-      value: 0.993140617344439
+      value: 0.7045565899069084
 ---
 
 <!-- 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: 7.8973
-- F1: 0.5024
-- Precision: 0.3362
-- Recall: 0.9931
+- Loss: 0.8029
+- F1: 0.6530
+- Precision: 0.6085
+- Recall: 0.7046
 - Support: None
 
 ## Model description
@@ -62,7 +62,7 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 3e-05
+- learning_rate: 2e-05
 - train_batch_size: 32
 - eval_batch_size: 32
 - seed: 42
@@ -78,8 +78,8 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | F1     | Precision | Recall | Support |
 |:-------------:|:-----:|:----:|:---------------:|:------:|:---------:|:------:|:-------:|
-| 6.0423        | 1.96  | 500  | 5.0312          | 0.5020 | 0.3358    | 0.9946 | None    |
-| 4.545         | 3.92  | 1000 | 7.8973          | 0.5024 | 0.3362    | 0.9931 | None    |
+| 2.4774        | 1.96  | 500  | 0.8233          | 0.6241 | 0.6118    | 0.6369 | None    |
+| 1.4697        | 3.92  | 1000 | 0.8029          | 0.6530 | 0.6085    | 0.7046 | None    |
 
 
 ### Framework versions

config.json

@@ -3,11 +3,11 @@
   "architectures": [
     "BertForSequenceClassification"
   ],
-  "attention_probs_dropout_prob": 0.5,
-  "classifier_dropout": 0.5,
+  "attention_probs_dropout_prob": 0.2,
+  "classifier_dropout": 0.2,
   "gradient_checkpointing": false,
   "hidden_act": "gelu",
-  "hidden_dropout_prob": 0.5,
+  "hidden_dropout_prob": 0.2,
   "hidden_size": 768,
   "id2label": {
     "0": "non-hate-speech",

config.toml

@@ -1,5 +1,5 @@
 [experiment]
-name = "binary-18"
+name = "binary-19"
 type = "binary"
 
 
@@ -13,9 +13,9 @@ augment_ratio = 0.0
 [model]
 pretrained_model_name_or_path = "thejosango/nuha-mlm"
 revision = "984ac09880b24959f6767fdbea8757d2c312df46"
-hidden_dropout_prob = 0.5
-attention_probs_dropout_prob = 0.5
-classifier_dropout = 0.5
+hidden_dropout_prob = 0.2
+attention_probs_dropout_prob = 0.2
+classifier_dropout = 0.2
 #num_hidden_layers = 6
 #num_attention_heads = 12
 #hidden_size = 768
@@ -26,7 +26,7 @@ classifier_dropout = 0.5
 num_train_epochs = 5
 warmup_steps = 1e3
 lr_scheduler_type = "constant"
-learning_rate = 3e-5
+learning_rate = 2e-5
 per_device_train_batch_size = 32 
 per_device_eval_batch_size = 32
 gradient_accumulation_steps = 4

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f2120e451b3207731f3ea2482404e5e855e15e0cdc2ec9d891e3cbc362bc96c
+oid sha256:fc1bc316b8ec21345ca7b68f4f5e1ff972ee277d8934a345c4f0e321bbdb6caf
 size 540847921

tokenizer.json

@@ -1,11 +1,6 @@
 {
   "version": "1.0",
-  "truncation": {
-    "direction": "Right",
-    "max_length": 512,
-    "strategy": "LongestFirst",
-    "stride": 0
-  },
+  "truncation": null,
   "padding": null,
   "added_tokens": [
     {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ad45f0d683378bbfdda4cdb0210bb80501d2ea5059e9e98d9f3d786101f9476
+oid sha256:b7678c7883253885dd4899d12b63604ea424b6184cc5ea0bc7dba8a4c6c4b86b
 size 4091