binary-52

README.md

@@ -24,13 +24,13 @@ model-index:
     metrics:
     - name: F1
       type: f1
-      value: 0.6726338729763388
+      value: 0.6902334758303189
     - name: Precision
       type: precision
-      value: 0.5811701412239408
+      value: 0.6218338023996445
     - name: Recall
       type: recall
-      value: 0.7982634398669869
+      value: 0.7755403657860706
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -40,10 +40,10 @@ 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: 0.5844
-- F1: 0.6726
-- Precision: 0.5812
-- Recall: 0.7983
+- Loss: 0.5773
+- F1: 0.6902
+- Precision: 0.6218
+- Recall: 0.7755
 - Support: None
 
 ## Model description
@@ -76,14 +76,14 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | F1     | Precision | Recall | Support |
 |:-------------:|:-----:|:----:|:---------------:|:------:|:---------:|:------:|:-------:|
-| 0.5778        | 0.37  | 500  | 0.5287          | 0.6244 | 0.6824    | 0.5755 | None    |
-| 0.5521        | 0.74  | 1000 | 0.5877          | 0.6650 | 0.5534    | 0.8330 | None    |
-| 0.5373        | 1.11  | 1500 | 0.5723          | 0.6741 | 0.5749    | 0.8147 | None    |
-| 0.5035        | 1.48  | 2000 | 0.5658          | 0.6712 | 0.6125    | 0.7425 | None    |
-| 0.5071        | 1.85  | 2500 | 0.5804          | 0.6737 | 0.5896    | 0.7857 | None    |
-| 0.4371        | 2.22  | 3000 | 0.5690          | 0.6666 | 0.6187    | 0.7225 | None    |
-| 0.4459        | 2.58  | 3500 | 0.5691          | 0.6631 | 0.6051    | 0.7334 | None    |
-| 0.4483        | 2.95  | 4000 | 0.5844          | 0.6726 | 0.5812    | 0.7983 | None    |
+| 0.5531        | 0.37  | 500  | 0.5324          | 0.6866 | 0.6190    | 0.7707 | None    |
+| 0.5346        | 0.74  | 1000 | 0.5675          | 0.6920 | 0.5862    | 0.8444 | None    |
+| 0.5142        | 1.11  | 1500 | 0.5444          | 0.6980 | 0.6032    | 0.8280 | None    |
+| 0.472         | 1.48  | 2000 | 0.5466          | 0.6993 | 0.6165    | 0.8079 | None    |
+| 0.4704        | 1.85  | 2500 | 0.5346          | 0.7020 | 0.6274    | 0.7968 | None    |
+| 0.3858        | 2.22  | 3000 | 0.5762          | 0.6989 | 0.6292    | 0.7861 | None    |
+| 0.3912        | 2.58  | 3500 | 0.5676          | 0.6929 | 0.6372    | 0.7593 | None    |
+| 0.3966        | 2.95  | 4000 | 0.5773          | 0.6902 | 0.6218    | 0.7755 | None    |
 
 
 ### Framework versions

config.json

@@ -31,5 +31,5 @@
   "transformers_version": "4.32.1",
   "type_vocab_size": 2,
   "use_cache": true,
-  "vocab_size": 17513
+  "vocab_size": 64000
 }

config.toml

@@ -1,5 +1,5 @@
 [experiment]
-name = "binary-51"
+name = "binary-52"
 type = "binary"
 
 
@@ -12,7 +12,7 @@ undersampling_strategy = true
 
 [model]
 pretrained_model_name_or_path = "thejosango/nuha-mlm"
-revision = "375b07db0be9afbc0412316eafd344f2cdbec1c8"
+revision = "ce20f497544665775129f9ff5b3cd2a3e350dce8"
 num_hidden_layers = 6
 classifier_dropout = 0.25
 
@@ -24,8 +24,8 @@ learning_rate = 5e-5
 per_device_train_batch_size = 64 
 per_device_eval_batch_size = 64
 gradient_accumulation_steps = 1
-weight_decay = 0.01
+weight_decay = 1e-3
 label_smoothing_factor = 0.1
-weighted_loss = false
+weighted_loss = false 
 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:a51ffcd7f300ce7ad7643241f6ea9e9824bb216d5a4b61061df6607c1fcdee1a
-size 227897969
+oid sha256:77854de095531e646b72ad7d6a9d3323b85a5652cee4d8a587558eaf19f4b5f4
+size 370706033

special_tokens_map.json

@@ -1,7 +1,4 @@
 {
-  "additional_special_tokens": [
-    "[URL]"
-  ],
   "cls_token": "[CLS]",
   "mask_token": "[MASK]",
   "pad_token": "[PAD]",

tokenizer_config.json

@@ -1,13 +1,17 @@
 {
-  "additional_special_tokens": [
-    "[URL]"
-  ],
   "clean_up_tokenization_spaces": true,
   "cls_token": "[CLS]",
-  "do_lower_case": true,
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
   "mask_token": "[MASK]",
+  "max_len": 512,
   "max_length": 512,
   "model_max_length": 512,
+  "never_split": [
+    "[بريد]",
+    "[مستخدم]",
+    "[رابط]"
+  ],
   "pad_to_multiple_of": null,
   "pad_token": "[PAD]",
   "pad_token_type_id": 0,
@@ -19,5 +23,6 @@
   "tokenizer_class": "BertTokenizer",
   "truncation_side": "right",
   "truncation_strategy": "longest_first",
-  "unk_token": "[UNK]"
+  "unk_token": "[UNK]",
+  "use_fast": true
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:902cd957b237c48acb38102c573c8dbb6fb7a90b64c987e7d6cd122063dc37af
+oid sha256:21f5bef3aca0c3c13753416b32c9243167e3d497cac8c70cb0b7b3354d422cb3
 size 4091