Multi-task Language Understanding on MMLU (Accuracy and vNMSE)
73.04AccuracyBF16
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BF16Communication Topology=butterfly all-reduce, Model=LLaMA 1B2026.02 | 73.04 | 0 | |
| DynamiQCommunication Topology=butterfly all-reduce, Model=LLaMA 1B2026.02 | 73.04 | 0.0007 | |
| MXFP8Communication Topology=butterfly all-reduce, Model=LLaMA 1B2026.02 | 72.86 | 0.002 | |
| MXFP6Communication Topology=butterfly all-reduce, Model=LLaMA 1B2026.02 | 72.46 | 0.0201 | |
| MXFP4Communication Topology=butterfly all-reduce, Model=LLaMA 1B2026.02 | 71.59 | 0.1706 |