backtesting-v11
Evaluation Results
| Metric | baseline_v11 |
|---|---|
| Evaluation Loss | 2.0789 |
| Perplexity (PPL) | 8.00 |
| BPC | 2.9992 |
v8 Improvements over v7
| Strategy | Description |
|---|---|
| Data Filtering | ํ๊ธ ๋น์จ 60%, ๋ฐ๋ณต์ค ํ์ง, ํน์๋ฌธ์ ๋น์จ ์ ํ |
| Text Cleaning | ์ํค ๋งํฌ์ /URL/HTML ์ ์ |
| Tokenizer | ByteLevel BPE (6k), ํ๊ธ ํ ํฐ๋ง ์ถ์ถ, PEFT ์์ resize |
| Curriculum Learning | ์งง์->๊ธด ๋ฌธ์ ์ ํ์ต์ผ๋ก ์ด๋ฐ ์์ ์ฑ ํฅ์ |
| Prompt Format | BOS ๋ช ์, ์์ฐ์ค๋ฌ์ด ํ๊ตญ์ด ์ํค ํฌ๋งท |
Experiment Details
- Version: baseline_v11
- Base Model: Llama-3.2-1B (Unsloth 4-bit)
- Korean BPE Tokens Added: 2662
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support
Model tree for zeronamoni/backtesting-v11
Base model
meta-llama/Llama-3.2-1B Quantized
unsloth/Llama-3.2-1B-unsloth-bnb-4bit