Table 2 Comparison of modern transformer models for zero-shot department classification.

From: A zero-shot LLM framework for multimodal grievance classification, urgency scoring, and abuse detection in civic feedback systems

Model

Accuracy (%)

Latency (ms)

GPU required

BERT-base (fine-tuned)

94.1

128

Yes

RoBERTa-base (fine-tuned)

94.8

141

Yes

DeBERTa-v3-base (fine-tuned)

95.2

158

Yes

LLaMA-3 8B (zero-shot)

93.7

420

Yes

MobileBERT (zero-shot)

92.4

19

No

Back to article page