Feature Space Preservation on WikiText-2
100Cosine SimilarityBF16
Evaluation Results
| Method | Links | |
|---|---|---|
| BF16Model Architecture=Llama3, Model Size=1B2026.03 | 100 | |
| BF16Model Architecture=Llama3, Model Size=8B2026.03 | 100 | |
| BF16Model Architecture=Qwen3, Model Size=1.7B2026.03 | 100 | |
| BF16Model Architecture=Qwen3, Model Size=8B2026.03 | 100 | |
| FAAR+2FAModel Architecture=Qwen3, Model Size=8B2026.03 | 99.13 | |
| MR-GPTQModel Architecture=Qwen3, Model Size=8B2026.03 | 99.08 | |
| FAAR+2FAModel Architecture=Qwen3, Model Size=1.7B2026.03 | 99.02 | |
| GPTQModel Architecture=Qwen3, Model Size=8B2026.03 | 99.01 | |
| FAAR+2FAModel Architecture=Llama3, Model Size=8B2026.03 | 98.96 | |
| 4/6Model Architecture=Qwen3, Model Size=8B2026.03 | 98.92 | |
| MR-GPTQModel Architecture=Qwen3, Model Size=1.7B2026.03 | 98.76 | |
| GPTQModel Architecture=Qwen3, Model Size=1.7B2026.03 | 98.75 | |
| GPTQ+4/6Model Architecture=Qwen3, Model Size=1.7B2026.03 | 98.68 | |
| GPTQ+4/6Model Architecture=Llama3, Model Size=8B2026.03 | 98.62 | |
| Strong BaselineModel Architecture=Qwen3, Model Size=8B2026.03 | 98.59 | |
| 4/6Model Architecture=Llama3, Model Size=8B2026.03 | 98.5 | |
| GPTQ+4/6Model Architecture=Qwen3, Model Size=8B2026.03 | 98.49 | |
| GPTQModel Architecture=Llama3, Model Size=8B2026.03 | 98.45 | |
| Strong BaselineModel Architecture=Qwen3, Model Size=1.7B2026.03 | 98.22 | |
| FAAR+2FAModel Architecture=Llama3, Model Size=1B2026.03 | 98.06 | |
| RTNModel Architecture=Qwen3, Model Size=1.7B2026.03 | 97.97 | |
| GPTQ+4/6Model Architecture=Llama3, Model Size=1B2026.03 | 97.93 | |
| RTNModel Architecture=Qwen3, Model Size=8B2026.03 | 97.92 | |
| MR-GPTQModel Architecture=Llama3, Model Size=1B2026.03 | 97.9 | |
| MR-GPTQModel Architecture=Llama3, Model Size=8B2026.03 | 97.8 | |
| GPTQModel Architecture=Llama3, Model Size=1B2026.03 | 97.78 | |
| Strong BaselineModel Architecture=Llama3, Model Size=8B2026.03 | 97.78 | |
| RTNModel Architecture=Llama3, Model Size=8B2026.03 | 97.23 | |
| 4/6Model Architecture=Qwen3, Model Size=1.7B2026.03 | 96.94 | |
| 4/6Model Architecture=Llama3, Model Size=1B2026.03 | 96.72 | |
| Strong BaselineModel Architecture=Llama3, Model Size=1B2026.03 | 96.55 | |
| RTNModel Architecture=Llama3, Model Size=1B2026.03 | 96.08 |