search

huggingface / optimum-amd

AMD related optimizations for transformer models
https://huggingface.co/docs/optimum/amd/index
MIT License
57 stars 17 forks source link

[RyzenAI] Operator regression on Ryzen SDK 1.1 #124

Open mht-sharma opened 7 months ago

mht-sharma commented 7 months ago

The following models have shown regression:

The format is: Ryzen1.1 value (Ryzen1.0.1 value)

Total Ops |    DPU Ops |   CPU Ops | Reg. | Model
      944 |   296(331) |  648(613) |    Y | timm/eca_botnext26ts_256.c1_in1k
     1212 | 1042(1126) |   170(86) |    Y | timm/eca_nfnet_l0.ra2_in1k
      755 |   613(683) |   142(72) |    Y | timm/eca_resnet33ts.ra2_in1k
      611 |   497(553) |   114(58) |    Y | timm/eca_resnext26ts.ch_in1k
      431 |   317(373) |   114(58) |    Y | timm/ecaresnet26t.ra2_in1k
     1325 | 1078(1131) |  247(194) |    Y | timm/gcresnet33ts.ra2_in1k
     1059 |   889(921) |  170(138) |    Y | timm/gcresnext26ts.ch_in1k