Closed Richsheep closed 2 years ago
Hi @Richsheep,
if you want get more info about the compiled model, you could use "xir" tool. One method of its use is as follows:
xir dump_txt
Hi @Richsheep, if you want get more info about the compiled model, you could use "xir" tool. One method of its use is as follows: xir dump_txt [] e.g. xir dump_txt a.xmodel a.txt when is missing, it dumps to standard output. the txt file would provide a lot of details. for code size, you can seach "mc_code", for parameter size, you can search "reg_id_to_size". but unfortunately, the workload info is not available until Vitis-AI-2.0.
Hi @wanghong4compiler Thanks,but still confusing why workload is removed in vitis-ai,and why models in model-zoo have workload marked.
Hi @Richsheep , sorry for the incorrect info, you can also get workload info. sorry again.
Hi @Richsheep , sorry for the incorrect info, you can also get workload info. sorry again.
Hi @wanghong4compiler How to get workload info? Have to find any other tools?
Hi @Richsheep , xir is a open source repo. you can get the source code from xilinx git. this depend on unilog repo, which is also a open source repo, and also in xilinx git. if you use the docker image, you can get xir directly without doing anything.
Hi @Richsheep , xir is a open source repo. you can get the source code from xilinx git. this depend on unilog repo, which is also a open source repo, and also in xilinx git. if you use the docker image, you can get xir directly without doing anything.
Many thanks, I found workload info in xir dump_txt output, it seems in subg_root
Thanks again for your support!
Hi, i also those info with the tool xir dump_txt . But i dont know which number represent Gops. What is the conversion relationship between them? I used the tool xir to generate the info of the model "tf2_efficientnet-b0_imagenet_224_224_0.36G_2.0" but the numbers here do not match it (0.36G)
thanks!
My question is the same. How do I find the detailed documentation for OPS? The information of dump_txt provided is not stated in any way!
Is there anyway or any tool to find out OPS of xmodels? Or only could calculate FLOPS of float model.
I remember in DNNDK 3.1, GOPS will shown at model compile result.
How can I get this param in Vitis-AI-1.3.2 for caffe/tensorflow compiled models?
thanks a lot.