Deployment Cost Analysis on General Queries
79.5Peak Memory (GB)SAG (10 agents)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SAG (10 agents)Backbone=Llama3.2-3B (Pre-trained)2026.02 | 79.5 | 28.6 | 63.9 | |
| SAG (10 agents)Backbone=Qwen3-4B (Pre-trained)2026.02 | 96.8 | 31.4 | 58.7 | |
| Llama-3-70BSystem=Single Agent (Large), Precision=4-bit quantization2026.02 | 132.5 | 23.7 | 27.8 | |
| Meditron-70BSystem=Clinical Single, Precision=4-bit quantization2026.02 | 138 | 23.3 | 32.6 | |
| Me-LLaMA-70BSystem=Clinical Single, Precision=4-bit quantization2026.02 | 141.2 | 22.1 | 38.9 | |
| Qwen-2.5-72BSystem=Single Agent (Large)2026.02 | 148.2 | 25.3 | 28.3 |