Medical Entity Extraction on Bangla MedER Entity V2 (test)
86.1AccuracyCRF-KD Student
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CRF-KD StudentModel Architecture=Student, Layers=4, Distillation=CRF-KD2026.05 | 86.1 | 40.17 | 49.29 | 44.56 | |
| BanglaBERT-CRFModel Architecture=Teacher, Layers=12, Head=CRF2026.05 | 85.5 | 39.62 | 49.68 | 47.86 | |
| TinyBanglaBERT (No-KD)Model Architecture=Student, Layers=4, Distillation=None2026.05 | 85.14 | 37.02 | 45.7 | 40.9 | |
| Quantized CRF-KDModel Architecture=Student, Layers=4, Distillation=CRF-KD, Precision=INT82026.05 | 85.13 | 38.51 | 46.67 | 42.2 | |
| Standard KD StudentModel Architecture=Student, Layers=4, Distillation=Standard KD2026.05 | 85.12 | 38.18 | 46.56 | 41.96 |