Named Entity Recognition on MTSamples (test) benchmark leaderboard · SOTA2 Research

Named Entity Recognition on MTSamples (test)

87Strict F1Centralized fine-tuning

Evaluation Results

Method
Centralized fine-tuning2026.01	87	95.8
LLaMA3-8B Centralized2026.01	86.7	92.6
Fed-MedLoRA+2026.01	86.6	93.7
Centralized fine-tuning2026.01	86.2	96.3
Fed-MedLoRA+2026.01	86.2	91.4
Fed-MedLoRA2026.01	86.1	90.1
Fed-MedLoRA+2026.01	86	95.1
Fed-MedLoRA2026.01	85.8	92.9
Fed-MedLoRA2026.01	85.8	93.6
DeepSeek-R1-Distill-8B Centralized2026.01	84.6	92.6
Single site fine-tuning2026.01	84.3	92.6
LLaMA3-8B Single site2026.01	84.1	86.7
Bio_ClinicalBERT Centralized2026.01	83.8	91.2
Single site fine-tuning2026.01	83.6	92.1
Centralized fine-tuning2026.01	82.7	90.5
DeepSeek-R1-Distill-8B Single site2026.01	82.5	85.5
Fed-MedLoRA+2026.01	81.8	92.5
Fed-MedLoRA2026.01	80.3	91
Single site fine-tuning2026.01	79.2	88.8
Bio_ClinicalBERT Single site2026.01	79	86.9
Zero-shot2026.01	60.2	83.4
GPT-4o Zero-shot2026.01	60.2	83.4
Zero-shot2026.01	43.7	64.4
LLaMA3-8B Zero-shot2026.01	43.7	64.4
Zero-shot2026.01	30.3	49.9
DeepSeek-R1-Distill-8B Zero-shot2026.01	30.3	49.9