Kernel Engineering on GPUMODE TriMul TTT-Discover setting
2,005.0307Median Latency (µs)EUREKAGENT-CUDA Graph
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EUREKAGENT-CUDA GraphLLM=GLM-5.1, Rank=12026.06 | 2,005.0307 | 2,014.1874 | |
| EUREKAGENT-INT8 BMMLLM=GLM-5.1, Rank=22026.06 | 2,006.9998 | 2,013.5141 | |
| EUREKAGENT-Fused Front-EndLLM=GLM-5.1, Rank=32026.06 | 2,016.5718 | 2,020.2674 | |
| EUREKAGENT-Triton AutotuneLLM=GLM-5.1, Rank=42026.06 | 2,030.6877 | 2,041.5578 | |
| josusamartinLLM=N/A, Rank=52026.06 | 2,096.0441 | 2,105.1655 | |
| TTT-DiscoverLLM=gpt-oss-120b, Rank=62026.06 | 2,247.7849 | 2,248.2307 | |
| rd9000LLM=N/A, Rank=72026.06 | 2,300.4883 | 2,307.5716 |