Knowledge augmentation on custom knowledge benchmark 250-question
85.8AccuracyPost-consolidation
Evaluation Results
| Method | Links | |
|---|---|---|
| Post-consolidationRetrieval Granularity=Section, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 85.8 | |
| Cold cacheRetrieval Granularity=Section, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 85.6 | |
| Warm cacheRetrieval Granularity=Section, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 85.4 | |
| Post-consolidationRetrieval Granularity=Section, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 85 | |
| Cold cacheRetrieval Granularity=Section, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 84 | |
| Warm cacheRetrieval Granularity=Section, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 84 | |
| Cold cacheRetrieval Granularity=Chunk, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 79.4 | |
| Warm cacheRetrieval Granularity=Chunk, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 78.8 | |
| Warm cacheRetrieval Granularity=Chunk, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 78.6 | |
| Cold cacheRetrieval Granularity=Chunk, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 78.2 | |
| Post-consolidationRetrieval Granularity=Chunk, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 78.2 | |
| Post-consolidationRetrieval Granularity=Chunk, Augmentation Mode=Augment, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 76.4 | |
| BaselineRetrieval Granularity=Section, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 33 | |
| BaselineRetrieval Granularity=Chunk, Augmentation Mode=Suppress, Threshold=0.80, Model=Qwen 3.5 2B2026.04 | 33 |