DACTYL Classifiers
Collection
Trained AI-generated text classifiers. Pretrained means using binary cross entropy loss, finetuned refers to deep X-risk optimized classifiers.
•
10 items
•
Updated
{
"training_split": "training",
"evaluation_split": "testing",
"results_path": "libauc-bert-tiny.csv",
"num_epochs": 1,
"model_path": "ShantanuT01/dactyl-bert-tiny-pretrained",
"tokenizer": "prajjwal1/bert-tiny",
"optimizer": "SOTAs",
"optimizer_type": "libauc",
"optimizer_args": {
"lr": 1e-05
},
"loss_fn": "tpAUC_KL_Loss",
"reset_classification_head": true,
"loss_type": "libauc",
"loss_fn_args": {
"data_len": 466005
},
"needs_loss_fn_as_parameter": false,
"save_path": "ShantanuT01/dactyl-bert-tiny-finetuned",
"training_args": {
"batch_size": 64,
"needs_sampler": true,
"needs_index": true,
"shuffle": false,
"sampling_rate": 0.5,
"apply_sigmoid": true
},
"best_model_path": "best-tpauc-model-bert-tiny"
}
| model | AP Score | AUC Score | OPAUC Score | TPAUC Score |
|---|---|---|---|---|
| DeepSeek-V3 | 0.986177 | 0.997357 | 0.991422 | 0.916469 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-RedditWritingPrompts-testing | 0.0242264 | 0.805738 | 0.619715 | 0 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-abstracts-testing | 0.473122 | 0.955828 | 0.857774 | 0.0525074 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-news-testing | 0.0223995 | 0.776301 | 0.579092 | 0 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-reviews-testing | 0.0167409 | 0.861497 | 0.526359 | 0 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-student_essays-testing | 0.0102312 | 0.621629 | 0.51884 | 0 |
| ShantanuT01/fine-tuned-Llama-3.2-1B-Instruct-apollo-mini-tweets-testing | 0.200169 | 0.949163 | 0.728122 | 0 |
| claude-3-5-haiku-20241022 | 0.96485 | 0.993769 | 0.977445 | 0.780686 |
| claude-3-5-sonnet-20241022 | 0.981121 | 0.997236 | 0.987999 | 0.883237 |
| gemini-1.5-flash | 0.93374 | 0.989888 | 0.955699 | 0.57296 |
| gemini-1.5-pro | 0.856575 | 0.96967 | 0.909663 | 0.190672 |
| gpt-4o-2024-11-20 | 0.950062 | 0.990051 | 0.968528 | 0.695722 |
| gpt-4o-mini | 0.992634 | 0.999149 | 0.995225 | 0.953515 |
| llama-3.2-90b | 0.895571 | 0.978201 | 0.933229 | 0.366601 |
| llama-3.3-70b | 0.952439 | 0.992252 | 0.968929 | 0.699379 |
| mistral-large-latest | 0.979345 | 0.996937 | 0.986758 | 0.871375 |
| mistral-small-latest | 0.986836 | 0.997499 | 0.991305 | 0.915284 |
| overall | 0.976784 | 0.980759 | 0.947371 | 0.492028 |