name,prefill_H100,prefill_MI300x,abs_diff Attention,5.848365577953972,18.518923695085483,12.670558117131511 Llama3RotaryEmbedding,1.7251192777954603,5.275438112527483,3.5503188347320225 LogitsProcessor,4.540207923847165,1.9657793266479366,-2.574428597199228 "MergedColumnParallelLinear(weight=bfloat16[28672, 4096])",37.7209583312918,31.706861989692143,-6.014096341599654 "QKVParallelLinear(weight=bfloat16[6144, 4096])",9.895869936136904,9.595384784728358,-0.30048515140854626 RMSNorm(weight=bfloat16[4096]) <- LlamaDecoderLayer,2.9142125957824905,3.1057451272984955,0.19153253151600502 RMSNorm(weight=bfloat16[4096]) <- LlamaForCausalLM,,0.049022471348246346, "RowParallelLinear(weight=bfloat16[4096, 14336])",23.992343837415707,19.238778404163504,-4.753565433252202 "RowParallelLinear(weight=bfloat16[4096, 4096])",8.607495036906135,6.0022756518021465,-2.6052193851039886 Sampler,1.565819497993548,1.5501282362977726,-0.015691261695775394 SiluAndMul,3.062675209068646,2.9435407394081348,-0.11913446966051122 others,0.04454403836961201,0.04812146100029267,0.003577422630680664 vocab_embed_ops,0.08238873743857462,,