Categorical Cross-Entropy Loss on CUDA-LLM task suite
6.4Time (ms)OptiML
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| OptiMLGPU=A100 80GB, Variant=G+X2026.02 | 6.4 | 1.08 | 37.94 | 25.31 | 31.46 | 900 | 2,272 | 1,657 | 80.22 | 46 | |
| HPC-Coder-V2 + OptiML-XGPU=A100 80GB2026.02 | 6.62 | 1.04 | 36.88 | 26.44 | 30.74 | 917 | 2,346 | 1,681 | 78.92 | 46 | |
| GPT 5.1 + OptiML-XGPU=A100 80GB2026.02 | 6.9 | — | 36.22 | 27.41 | 30.18 | 944 | 2,412 | 1,713 | 78.11 | 46 | |
| StarCoder2 + OptiML-XGPU=A100 80GB2026.02 | 7.08 | 0.97 | 35.14 | 27.86 | 29.62 | 956 | 2,451 | 1,732 | 77.03 | 46 | |
| Qwen2.5-Coder + OptiML-XGPU=A100 80GB2026.02 | 7.35 | 0.94 | 34.66 | 28.62 | 29.04 | 979 | 2,514 | 1,809 | 76.24 | 46 |