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
| { | |
| "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." | |
| } | |