name,decode_1_H100,decode_1_MI300x,abs_diff Attention,10.431836463519645,7.674151678857323,-2.757684784662321 Llama3RotaryEmbedding,1.7333178765562836,2.1757345407444877,0.44241666418820413 LogitsProcessor,4.931779492636019,4.584755769052253,-0.34702372358376543 "MergedColumnParallelLinear(weight=bfloat16[28672, 4096])",36.53517257642745,35.7003415580844,-0.8348310183430456 "QKVParallelLinear(weight=bfloat16[6144, 4096])",9.450435047807359,9.746143482157347,0.2957084343499883 RMSNorm(weight=bfloat16[4096]) <- LlamaDecoderLayer,2.9514035528290163,5.235026143898585,2.2836225910695687 RMSNorm(weight=bfloat16[4096]) <- LlamaForCausalLM,0.04489645754478734,0.07783741765778085,0.03294096011299351 "RowParallelLinear(weight=bfloat16[4096, 14336])",19.83281171434743,19.1229496321115,-0.7098620822359294 "RowParallelLinear(weight=bfloat16[4096, 4096])",8.319421289197242,7.666690669139791,-0.6527306200574516 Sampler,1.7282585219655353,3.2277195597310224,1.4994610377654871 SiluAndMul,3.9984915470514317,4.719852098495739,0.7213605514443073 others,0.04217546011783053,0.06879745006979515,0.026621989951964624