| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Qwen3 family S=512 (train profiling) | FORGE | Per-Step Latency (ms)100.2 | 43 | 1mo ago | |
| Qwen3 family S=1024 (train profiling) | FORGE | Per-Step Latency (ms)100.6 | 42 | 1mo ago | |
| Qwen3 profiling S=2048 (train) | FORGE | Per-Step Latency (ms)104.1 | 40 | 1mo ago | |
| Qwen3 profiling S=4096 (train) | FORGE | Per-step Latency (ms)108 | 39 | 1mo ago | |
| Qwen3 (train) | AdaLomo | Peak GPU Memory3.1 | 31 | 1mo ago |