Summarization on XSum (ROUGE-L F1)
18.87ROUGE-L F1LoRA FP16
Evaluation Results
| Method | Links | |
|---|---|---|
| LoRA FP16Rank=r16, BPW=16.0, Size (MB)=76.3, Time=–2026.05 | 18.87 | |
| BitDelta-col+distillRank=r16, BPW=1.00, Size (MB)=4.8, Time=+17 min2026.05 | 18.7 | |
| QAT FullRank=r64, BPW=1.0, Size (MB)=19.1, Time=+15h2026.05 | 18.6 | |
| QAT FullRank=r16, BPW=2.0, Size (MB)=9.5, Time=+11h2026.05 | 18.49 | |
| QAT FullRank=r16, BPW=4.0, Size (MB)=19.1, Time=+11h2026.05 | 18.49 | |
| QAT FullRank=r16, BPW=1.0, Size (MB)=7.2, Time=+11h2026.05 | 18.32 | |
| LoRAQuant ρ=0.9Rank=r16, BPW=1.68, Size (MB)=8.0, Time=+55 min2026.05 | 18.27 | |
| LoRAQuant ρ=0.5Rank=r16, BPW=1.20, Size (MB)=5.7, Time=+36 min2026.05 | 17.67 | |
| PTQ-LoRDBARank=r16, BPW=1.0, Size (MB)=7.2, Time=+2 min2026.05 | 7.24 |