BRlkl commited on
Commit
e16ce79
·
verified ·
1 Parent(s): 3dc3419

Upload benchmark_results.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. benchmark_results.md +24 -0
benchmark_results.md ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Hyperparameter Search Results for `BRlkl/BingoGuard-bert-large-pt-optimized`
2
+
3
+ This model was selected through a randomized search over a hyperparameter space.
4
+ The best performing model was chosen based on the highest average F1 score across multiple benchmark datasets.
5
+
6
+ ## Best Hyperparameters
7
+
8
+ ```
9
+ learning_rate: 5e-05
10
+ per_device_train_batch_size: 32
11
+ num_train_epochs: 8
12
+ weight_decay: 0.05
13
+ lr_scheduler_type: cosine
14
+ ```
15
+
16
+ ## Final Benchmark Results (Full Dataset Training)
17
+
18
+ | dataset | accuracy | f1_score | recall | precision |
19
+ |:----------------------------------|-----------:|-----------:|---------:|------------:|
20
+ | BRlkl/BingoGuard-train-test-pt | 0.907895 | 0.951724 | 0.907895 | 1 |
21
+ | BRlkl/openai-moderation-eval-pt | 0.684524 | 0.646195 | 0.927203 | 0.495902 |
22
+ | BRlkl/WildGuardTest-pt | 0.845792 | 0.81856 | 0.78382 | 0.856522 |
23
+ | BRlkl/XSTest-pt | 0.851111 | 0.839329 | 0.875 | 0.806452 |
24
+ | BRlkl/toxic-chat-pt (40% holdout) | 0.95821 | 0.763231 | 0.901316 | 0.661836 |