| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Long sequence prompts | FedRAG | TTFT (s)1.05 | 24 | 2mo ago | |
| Sequence prompts Medium | FedRAG | TTFT (s)0.37 | 24 | 2mo ago | |
| Short sequence prompts | FedRAG | TTFT (s)0.2 | 24 | 2mo ago | |
| AIME | RelayCaching | Time To First Token (TTFT)40.6 | 12 | 4mo ago | |
| Uniform Sampled Documents | Lazy-Attn | TTFT (ms)191.7 | 9 | 1mo ago | |
| HarmBench-Standard and AdvBench | ALIGNBEAM | Slowdown2 | 8 | 1mo ago | |
| Instruction-in-Wild self-instructed prompts 36 | Throughput (samples/s)2.52 | 7 | 4mo ago | ||
| LongBench-E | Prefill Time (sec)3.032 | 5 | 4mo ago | ||
| Edge Device | RAP | Latency (s)43.36 | 4 | 2mo ago | |
| Synthetic LLM Workload Input 4096 Output 4096 | ChunkKV | Latency (s)160.32 | 4 | 2mo ago | |
| Efficient Qwen Competition Scenarios Short (64/128), Medium (2048/256), Long (8192/256) | AFM-k5984d3s (Ours) | Average Speedup6.978 | 2 | 18d ago | |
| HumanEval | ReMoE | TTFT (ms)5,233.11 | 2 | 1mo ago | |
| Synthetic LLM Workload Input 8192 Output 4096 | ShotKV | Latency (s)162.78 | 2 | 2mo ago | |
| Self-instructed prompts (test) | - | Throughput (samples/s)- | 0 | 4mo ago |
