Intent Classification on CLINC-150 and 20 Newsgroups (test)
67.8Baseline AccuracyQwen2.5-32B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2.5-32BParameters=32B, Framework=Baseline2025.12 | 67.8 | — | — | — | — | |
| Qwen3-4BParameters=4B, Framework=Baseline2025.12 | 65.2 | — | — | — | — | |
| Qwen2.5-3BParameters=3B, Framework=Baseline2025.12 | 64.1 | — | — | — | — | |
| Qwen3-1.7BParameters=1.7B, Framework=Baseline2025.12 | 62.3 | — | — | — | — | |
| Qwen2.5-32BParameters=32B, Framework=Hybrid (Deep Alignment)2025.12 | — | 81.5 | 13.7 | 90.3 | 12 | |
| Qwen2.5-3BParameters=3B, Framework=Hybrid (Deep Alignment)2025.12 | — | 78.3 | 14.2 | 88.4 | 9 | |
| Qwen3-1.7BParameters=1.7B, Framework=Hybrid (Deep Alignment)2025.12 | — | 76.4 | 14.1 | 87.4 | 8 | |
| Qwen3-4BParameters=4B, Framework=Hybrid (Deep Alignment)2025.12 | — | 79.1 | 13.9 | 89.2 | 10 |