applying O3 quant config to llama

[cjm715]

how can I go about applying o3 quant config (static per-tensor a8w8) to llama? I don't see this in example notebooks

[cjm715]

In tables 6 and 7 of paper with smoothquant results for llama, which smoothquant config was used? O1, O2, or O3? I'm especially curious if static quantization was ever applied to llama in papers results