| language: ko | |
| license: apache-2.0 | |
| base_model: unsloth/Llama-3.2-1B-unsloth-bnb-4bit | |
| tags: | |
| - continued-pretraining | |
| - wikipedia | |
| # backtesting-base_b1 Comparison Table | |
| ## Evaluation Results | |
| | Metric | v2_branch_1 | | |
| | --- | --- | | |
| | Evaluation Loss | 2.6116 | | |
| | Perplexity (PPL) | 13.62 | | |
| | BPC | 3.7677 | | |
| ## Experiment Details | |
| * **Version**: v2_branch_1 | |
| * **Method**: train dataset 필터링 | |
| * **Base Model**: Llama-3.2-1B (Unsloth 4-bit) | |