---
language: ko
license: apache-2.0
base_model: unsloth/Llama-3.2-1B-unsloth-bnb-4bit
tags:
- continued-pretraining
- wikipedia
---

# backtesting-v6 Comparison Table

## Evaluation Results
| Metric | baseline_v6 |
| --- | --- |
| Evaluation Loss | 2.1891 |
| Perplexity (PPL) | 8.93 |
| BPC | 3.1583 |

## Experiment Details
* **Version**: baseline_v6
* **Method**: 600자 이상 데이터 필터링 및 15k Vocab 적용
* **Base Model**: Llama-3.2-1B (Unsloth 4-bit)