Question Answering on ARC-C (Accuracy, Throughput)
129Throughput (tokens/s)BICACHE + Fast-dLLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BICACHE + Fast-dLLMSetting=Integration Results (§6.3)2026.05 | 129 | 84.3 | |
| BICACHE + Fast-dLLMcaching=BICACHE, optimization=Fast-dLLM2026.05 | 123 | 84.7 | |
| vLLMSetting=Main Improvements (§6.2)2026.05 | 113.3 | 0 | |
| BICACHESetting=Main Improvements (§6.2)2026.05 | 111.7 | 85.3 | |
| vLLMframework=vLLM2026.05 | 90.61 | 0 | |
| BICACHEcaching=BICACHE2026.05 | 89.8 | 85.1 | |
| Fast-dLLMSetting=Integration Results (§6.3)2026.05 | 85.1 | 84.3 | |
| Fast-dLLMoptimization=Fast-dLLM2026.05 | 79.8 | 84.4 | |
| No cachingSetting=Main Improvements (§6.2)2026.05 | 79.4 | 86.1 | |
| No cachingcaching=None2026.05 | 65.7 | 86.2 |