Throughput Measurement on System Throughput Evaluation (test)
6,898.53Throughput (tokens/GPU/sec)GKA
Evaluation Results
| Method | Links | |
|---|---|---|
| GKAScale=8B, Context Length=16K, Batch size=4M tokens2025.11 | 6,898.53 | |
| GKAScale=8B, Context Length=32K, Batch size=4M tokens2025.11 | 6,808.94 | |
| GKAScale=8B, Context Length=64K, Batch size=4M tokens2025.11 | 6,721.64 | |
| TransformerScale=8B, Context Length=16K, Batch size=4M tokens2025.11 | 6,096.37 | |
| TransformerScale=8B, Context Length=32K, Batch size=4M tokens2025.11 | 4,766.25 | |
| TransformerScale=8B, Context Length=64K, Batch size=4M tokens2025.11 | 3,382.5 | |
| GKAScale=14B, Context Length=16K, Batch size=4M tokens2025.11 | 3,318.28 | |
| GKAScale=14B, Context Length=32K, Batch size=4M tokens2025.11 | 3,256.45 | |
| GKAScale=14B, Context Length=64K, Batch size=4M tokens2025.11 | 3,256.45 | |
| TransformerScale=14B, Context Length=16K, Batch size=4M tokens2025.11 | 2,833.99 | |
| TransformerScale=14B, Context Length=32K, Batch size=4M tokens2025.11 | 2,221.56 | |
| TransformerScale=14B, Context Length=64K, Batch size=4M tokens2025.11 | 1,485.24 |