Multi-turn Dialogue on MTBench101
9.03ScoreFull-Context
Evaluation Results
| Method | Links | |
|---|---|---|
| Full-ContextModel=Qwen3-4B, Compression Ratio=-2026.01 | 9.03 | |
| SONICModel=Qwen3-4B, Compression Ratio=50%2026.01 | 8.43 | |
| SONICModel=Qwen3-4B, Compression Ratio=80%2026.01 | 8.3 | |
| Full-ContextModel=Qwen3-1.7B, Compression Ratio=-2026.01 | 8.11 | |
| SONICModel=Qwen3-1.7B, Compression Ratio=50%2026.01 | 6.42 | |
| SONICModel=Qwen3-1.7B, Compression Ratio=80%2026.01 | 6.32 | |
| Full-ContextModel=Qwen3-0.6B, Compression Ratio=-2026.01 | 6 | |
| SONICModel=Qwen3-0.6B, Compression Ratio=50%2026.01 | 5.61 | |
| ExpectedAttentionModel=Qwen3-4B, Compression Ratio=50%2026.01 | 5.06 | |
| H2OModel=Qwen3-4B, Compression Ratio=50%2026.01 | 5.04 | |
| ExpectedAttentionModel=Qwen3-4B, Compression Ratio=80%2026.01 | 5.02 | |
| H2OModel=Qwen3-4B, Compression Ratio=80%2026.01 | 5 | |
| SnapKVModel=Qwen3-4B, Compression Ratio=50%2026.01 | 5 | |
| StreamingLLMModel=Qwen3-4B, Compression Ratio=50%2026.01 | 4.98 | |
| H2OModel=Qwen3-1.7B, Compression Ratio=50%2026.01 | 4.88 | |
| ExpectedAttentionModel=Qwen3-1.7B, Compression Ratio=50%2026.01 | 4.84 | |
| SnapKVModel=Qwen3-1.7B, Compression Ratio=50%2026.01 | 4.76 | |
| ExpectedAttentionModel=Qwen3-1.7B, Compression Ratio=80%2026.01 | 4.74 | |
| H2OModel=Qwen3-1.7B, Compression Ratio=80%2026.01 | 4.74 | |
| StreamingLLMModel=Qwen3-1.7B, Compression Ratio=50%2026.01 | 4.71 | |
| SnapKVModel=Qwen3-4B, Compression Ratio=80%2026.01 | 4.31 | |
| SnapKVModel=Qwen3-1.7B, Compression Ratio=80%2026.01 | 4.21 | |
| StreamingLLMModel=Qwen3-4B, Compression Ratio=80%2026.01 | 4.11 | |
| SONICModel=Qwen3-0.6B, Compression Ratio=80%2026.01 | 3.93 | |
| SnapKVModel=Qwen3-0.6B, Compression Ratio=50%2026.01 | 3.93 | |
| ExpectedAttentionModel=Qwen3-0.6B, Compression Ratio=50%2026.01 | 3.92 | |
| H2OModel=Qwen3-0.6B, Compression Ratio=50%2026.01 | 3.92 | |
| StreamingLLMModel=Qwen3-1.7B, Compression Ratio=80%2026.01 | 3.77 | |
| H2OModel=Qwen3-0.6B, Compression Ratio=80%2026.01 | 3.66 | |
| ExpectedAttentionModel=Qwen3-0.6B, Compression Ratio=80%2026.01 | 3.62 | |
| StreamingLLMModel=Qwen3-0.6B, Compression Ratio=50%2026.01 | 3.58 | |
| SnapKVModel=Qwen3-0.6B, Compression Ratio=80%2026.01 | 3.21 | |
| StreamingLLMModel=Qwen3-0.6B, Compression Ratio=80%2026.01 | 2.74 |