Initial training on seed20; DeBERTa-v3 toxicity classifier.

README.md

@@ -21,12 +21,12 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [microsoft/deberta-v3-base](https://huggingface.co/microsoft/deberta-v3-base) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.3717
-- Accuracy: 0.8059
-- Precision: 0.7445
-- Recall: 0.9940
-- F1: 0.8514
-- Auc: 0.8921
+- Loss: 0.3694
+- Accuracy: 0.8054
+- Precision: 0.7440
+- Recall: 0.9942
+- F1: 0.8511
+- Auc: 0.8908
 
 ## Model description
 
@@ -47,7 +47,7 @@ More information needed
 The following hyperparameters were used during training:
 - learning_rate: 5e-05
 - train_batch_size: 32
-- eval_batch_size: 16
+- eval_batch_size: 8
 - seed: 13
 - gradient_accumulation_steps: 8
 - total_train_batch_size: 256
@@ -59,19 +59,19 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1     | Auc    |
 |:-------------:|:-----:|:----:|:---------------:|:--------:|:---------:|:------:|:------:|:------:|
-| No log        | 1.0   | 141  | 0.4034          | 0.8042   | 0.7511    | 0.9722 | 0.8474 | 0.8866 |
-| No log        | 2.0   | 282  | 0.3723          | 0.8034   | 0.7462    | 0.9829 | 0.8483 | 0.8835 |
-| No log        | 3.0   | 423  | 0.3679          | 0.8044   | 0.7432    | 0.9938 | 0.8504 | 0.8905 |
-| 0.4243        | 4.0   | 564  | 0.3718          | 0.8059   | 0.7445    | 0.9940 | 0.8514 | 0.8921 |
-| 0.4243        | 5.0   | 705  | 0.3872          | 0.8053   | 0.7444    | 0.9928 | 0.8509 | 0.8918 |
-| 0.4243        | 6.0   | 846  | 0.3934          | 0.8050   | 0.7453    | 0.9895 | 0.8502 | 0.8917 |
-| 0.4243        | 7.0   | 987  | 0.3964          | 0.8050   | 0.7480    | 0.9823 | 0.8493 | 0.8914 |
-| 0.3214        | 8.0   | 1128 | 0.3980          | 0.8066   | 0.7557    | 0.9668 | 0.8484 | 0.8918 |
+| No log        | 1.0   | 141  | 0.4441          | 0.8012   | 0.7428    | 0.9861 | 0.8473 | 0.8880 |
+| No log        | 2.0   | 282  | 0.3568          | 0.8042   | 0.7453    | 0.9875 | 0.8495 | 0.8905 |
+| No log        | 3.0   | 423  | 0.3691          | 0.8052   | 0.7444    | 0.9926 | 0.8508 | 0.8922 |
+| 0.4062        | 4.0   | 564  | 0.3701          | 0.8054   | 0.7440    | 0.9942 | 0.8511 | 0.8908 |
+| 0.4062        | 5.0   | 705  | 0.3925          | 0.8051   | 0.7436    | 0.9944 | 0.8509 | 0.8915 |
+| 0.4062        | 6.0   | 846  | 0.3891          | 0.8056   | 0.7498    | 0.9793 | 0.8493 | 0.8921 |
+| 0.4062        | 7.0   | 987  | 0.3860          | 0.8070   | 0.7573    | 0.9638 | 0.8482 | 0.8943 |
+| 0.3208        | 8.0   | 1128 | 0.3909          | 0.8073   | 0.7603    | 0.9575 | 0.8475 | 0.8939 |
 
 
 ### Framework versions
 
 - Transformers 4.57.1
-- Pytorch 2.9.1+cu130
+- Pytorch 2.8.0+cu129
 - Datasets 4.4.1
 - Tokenizers 0.22.1

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d99db14f2644bfd06f25d635b61256a22acc62f23590f7478afa980a49b91295
+oid sha256:49df3870ca7de2415f1c78353313e1c28f3ada9129d2e1fb58df386ad4cc8556
 size 737719272

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05a5387fe13c91ba49c02ee1595fe39e5ed4db343f288461d69d13b01fa3dd96
+oid sha256:288663d4dae564292a49ddcd99d203282db760c60bcffd017f24a988a8d2a398
 size 5841