Instructions to use KETI-NLP/KoEVD-assistant-safety with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use KETI-NLP/KoEVD-assistant-safety with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-0.8B") model = PeftModel.from_pretrained(base_model, "KETI-NLP/KoEVD-assistant-safety") - Notebooks
- Google Colab
- Kaggle
File size: 812 Bytes
7c3fc28 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | {
"evaluation_set": "assistant_safety_source_disjoint_eval_532",
"num_source_ids": 100,
"num_examples": 532,
"source_disjoint_from_training": true,
"true_label_distribution": {
"safe": 232,
"unsafe": 300
},
"predicted_label_distribution": {
"safe": 233,
"unsafe": 299
},
"num_parse_errors": 0,
"accuracy": 0.9981203007518797,
"micro_f1": 0.9981203007518797,
"macro_f1": 0.9980900066418943,
"per_label_f1": {
"safe": 0.9978494623655914,
"unsafe": 0.9983305509181971
},
"confusion_matrix": {
"rows": ["safe", "unsafe"],
"columns": ["safe", "unsafe"],
"counts": [[232, 0], [1, 299]]
},
"base_model": "Qwen/Qwen3.5-0.8B",
"release_note": "Sanitized source-disjoint validation metrics; local paths, runtime, and hardware identifiers removed."
}
|