Four-class classification on UDC (evaluation)
0.674Precision (%)Issue discovery agent
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Issue discovery agentTools=Tools 1 & 2, Base Model=REDA 7B2026.01 | 0.674 | 0.5792 | 0.9089 | 0.5883 | |
| Issue discovery agentTools=Tools 1 & 3, Base Model=REDA 7B2026.01 | 0.52 | 0.4675 | 0.8977 | 0.4736 | |
| Issue discovery agentTools=Tool 1 (governance models), Base Model=REDA 7B2026.01 | 0.5102 | 0.4917 | 0.9031 | 0.4709 | |
| Issue discovery agentTools=No tools, Base Model=REDA 7B2026.01 | 0.5041 | 0.459 | 0.8977 | 0.4575 | |
| Issue discovery agentTools=Tool 2 (embedding models), Base Model=REDA 7B2026.01 | 0.4965 | 0.5225 | 0.8927 | 0.4856 | |
| Issue discovery agentTools=Tool 3 (long-term memory), Base Model=REDA 7B2026.01 | 0.4946 | 0.5185 | 0.8923 | 0.483 | |
| Supervised fine-tuningBase Model=REDA 7B, Tuning=LoRA2026.01 | 0.4587 | 0.3371 | 0.9073 | 0.3671 | |
| Supervised fine-tuningBase Model=Qwen2.5-VL 7B, Tuning=LoRA2026.01 | 0.4023 | 0.3213 | 0.9024 | 0.3419 |