LLM Training on Llama2-7B tpu-v5p-512
2.5Iteration Time (s)AXLearn
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AXLearnHardware=tpu-v5p-5122025.07 | 2.5 | 66.2 | — | |
| MaxTextHardware=tpu-v5p-5122025.07 | 2.7 | 61.6 | — | |
| PyTorch XLA FSDPHardware=tpu-v5p-5122025.07 | 3.5 | 46.7 | — |
Loading the SOTA2 catalog…