roofAI 🏛️ on GPU — 2

Next, in roofAI 🏛️, boot up SageMaker with GPU,

from https://aws.amazon.com/sagemaker/pricing/

@seed sagemaker plugin=roofAI,screen

on Mac,

⚙️ python3 -m roofAI.semseg predict — device cpu — model_path /Users/kamangir/storage/abcli/2023–10–29–12–57–16–38247 — dataset_path /Users/kamangir/storage/abcli/roofAI_ingest_AIRS_2023–11–12–19–44–22–27771 — prediction_path /Users/kamangir/storage/abcli/2023–11–26–19–22–40–31128 — profile FULL

+= tqdm.

why doesn’t vscode click not work? fixed, thanks to OpenAI.

report prediction time.

took 390 millisecond(s)

roofAI semseg predict \
    profile=VALIDATION \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

roofAI semseg predict \
    device=cpu,profile=VALIDATION \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

took 333 millisecond(s)

validate a roofAI predict on GPU,

roofAI semseg predict \
    device=cuda,profile=VALIDATION \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

pr_mask = self.model.predict(x_tensor)

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

using https://stackoverflow.com/questions/59013109/runtimeerror-input-type-torch-floattensor-and-weight-type-torch-cuda-floatte

took 235 millisecond(s)

roofAI semseg predict \
    device=cpu,profile=QUICK \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

roofAI semseg predict \
    device=cuda,profile=QUICK \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

10+ times faster. 🚀

duplicate item in the roofAI signature, investigate ✅ document elapsed time ✅ number of objects on the image ✅

roofAI semseg predict \
    device=cpu,profile=VALIDATION \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

roofAI semseg predict \
    profile=VALIDATION \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

roofAI semseg predict \
    profile=QUICK \
    $(@ref roofAI_semseg_model_AIRS_full_v2) \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp)

validate train on CPU ✅ on GPU ✅

roofAI semseg train \
    profile=VALIDATION \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp) \
    --classes roof

RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

pr_mask = self.model.predict(x_tensor)

roofAI semseg train \
    profile=QUICK,upload \
    $(@ref roofAI_ingest_AIRS_v2) \
    $(@timestamp) \
    --classes roof

Next: training on GPU.