Dialogue State Tracking on Public (test)
97.98Joint Goal Accuracy (JGA)+ Stream-Only
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| + Stream-OnlyBackbone=Gemma3-4B2026.05 | 97.98 | 99.49 | |
| + StreamDial (Hybrid)Backbone=Gemma3-4B2026.05 | 97.98 | 99.54 | |
| + Seed-OnlyBackbone=Gemma3-4B2026.05 | 97.73 | 99.49 | |
| + StreamDial (Hybrid)Backbone=Qwen3-1.7B2026.05 | 96.97 | 99.37 | |
| + StreamDial (Hybrid)Backbone=Qwen3-8B2026.05 | 96.72 | 99.41 | |
| + Stream-OnlyBackbone=Qwen3-1.7B2026.05 | 96.46 | 99.32 | |
| + Stream-OnlyBackbone=Qwen3-8B2026.05 | 95.96 | 99.12 | |
| + Seed-OnlyBackbone=Qwen3-8B2026.05 | 95.71 | 99.2 | |
| SFT w/ Public Data (Baseline)Backbone=Gemma3-4B2026.05 | 95.71 | 98.78 | |
| + Seed-OnlyBackbone=Qwen3-1.7B2026.05 | 95.45 | 98.82 | |
| SFT w/ Public Data (Baseline)Backbone=Qwen3-8B2026.05 | 93.94 | 98.56 | |
| SFT w/ Public Data (Baseline)Backbone=Qwen3-1.7B2026.05 | 93.43 | 98.81 | |
| Zero-shot (No SFT)Backbone=Qwen3-8B2026.05 | 54.04 | 73.4 | |
| Zero-shot (No SFT)Backbone=Qwen3-1.7B2026.05 | 45.71 | 74.61 | |
| Zero-shot (No SFT)Backbone=Gemma3-4B2026.05 | 33.59 | 67.93 |