Qualcomm Aimet Efficiency Toolkit Tuhinga Tohutohu

KBA-231226181840

1. Tatūnga Taiao

1.1. Tāutahia te taraiwa Nvidia me te CUDA

1.2. Tāutahia te Whare Pukapuka Python e Pa ana

python3 -m pip tāuta -whakahou -kaore e whakauruhia te pip
python3 -m pip tāuta –ignore-installed gdown
python3 -m pip tāuta -kaore i whakauruhia opencv-python
python3 -m pip tāuta –ignore-installed torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
python3 -m pip tāuta –ignore-installed jax
python3 -m pip tāuta –ignore-installed ftfy
python3 -m pip tāuta –whakaarohia te torchinfo kua whakauruhia
python3 -m pip tāuta –ignore-installed https://github.com/quic/aimet/releases/download/1.25.0/AimetCommon-torch_gpu_1.25.0-cp38-cp38-linux_x86_64.whl
python3 -m pip tāuta –ignore-installed https://github.com/quic/aimet/releases/download/1.25.0/AimetTorch-torch_gpu_1.25.0-cp38-cp38-linux_x86_64.whl
python3 -m pip tāuta –ignore-installed numpy ==1.21.6
python3 -m pip tāuta -kaore i whakauruhia te psutil

1.3. Clone aimet-model-zoo

git clone https://github.com/quic/aimet-model-zoo.git
cd aimet-model-zoo
git checkout d09d2b0404d10f71a7640a87e9d5e5257b028802
kaweake PYTHONPATH=${PYTHONPATH}:${PWD}

1.4. Tikiake Set14

wget https://uofi.box.com/shared/static/igsnfieh4lz68l926l8xbklwsnnk8we9.zip
unzip igsnfieh4lz68l926l8xbklwsnnk8we9.zip

1.5. Whakakē raina 39 aimet-model-zoo/aimet_zoo_torch/quicksrnet/dataloader/utils.py

huri
mo img_path i glob.glob(os.path.join(test_images_dir, “*”)):
ki
mo img_path i glob.glob(os.path.join(test_images_dir, “*_HR.*”)):

1.6. Whakahaerehia te aro mātai.

# rere i raro i YOURPATH/aimet-model-run
# Mo te quicksrnet_small_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_small_2x_w8a8 \
–raunga-raunga ../Set14/image_SRF_4

# Mo te quicksrnet_small_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_small_4x_w8a8 \
–raunga-raunga ../Set14/image_SRF_4

# Mo te quicksrnet_medium_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_medium_2x_w8a8 \
–raunga-raunga ../Set14/image_SRF_4

# Mo te quicksrnet_medium_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_medium_4x_w8a8 \
–raunga-raunga ../Set14/image_SRF_4

me whakaaro ka whiwhi koe i te uara PSNR mo te tauira whakatairite. Ka taea e koe te whakarereke i te tauira-whirihora mo te rahinga rereke oQuickSRNet, ko te whiringa ko te underaimet-modelzoo/aimet_zoo_torch/quicksrnet/model/model_cards/.

2 Tāpiri Papaki

2.1. Whakatuwheratia "Kaweake ki ONNX Steps REVISED.docx"

2.2. Tīpoka git commit id

2.3. Waehere Wāhanga 1

Tāpirihia te 1. waehere katoa i raro i te raina whakamutunga (i muri i te raina 366) aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/models.py

2.4. Wāhanga 2 me 3 Waehere

Tāpirihia te 2, 3 waehere katoa i raro i te rarangi 93 aimet-model-zoo/aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py

2.5. Tawhā Matua i roto i te Tauira uta_mahi

tauira = uta_tauira(MODEL_PATH_INT8,

MODEL_NAME,
MODEL_ARGS. tiki(MODEL_NAME). tiki(MODEL_CONFIG),
use_quant_sim_model=Tono,
encoding_path=ENCODING_PATH,
quantsim_config_path=CONFIG_PATH,
calibration_data=IMAGES_LR,
use_cuda=Tono,
before_quantization=Tono,
convert_to_dcr=Tono)

MODEL_PATH_INT8 = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/pre_opt_weights
MODEL_NAME = TereSRNetSmall
MODEL_ARGS. tiki(MODEL_NAME). tiki(MODEL_CONFIG) = {'tauine_factor': 2}
ENCODING_PATH = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/adaround_encodings
CONFIG_PATH = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/aimet_config

Tena koa whakakapihia nga taurangi mo nga rahi rereke o QuickSRNet

2.6 Tauira Rahi Whakarerekētanga

1. "whakauru_ahua" i roto i te aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/model_cards/*.json
2. I roto i te mahi load_model(…) i roto i te aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/inference.py
3. Tawhā o roto mahi export_to_onnx(…, tāuru_teitei, tāuru_whānui) mai i “Kaweake ki ONNX Steps REVISED.docx”

2.7 Re-Run 1.6 ano mo te kaweake tauira ONNX

3. Tahuri ki SNPE

3.1. Tahuri

${SNPE_ROOT}/bin/x86_64-linux-clang/snpe-onnx-to-dlc \
–input_network model.onnx \
–quantization_overrides ./model.encodings

3.2. (Kōwhiringa) Tangohia anake te DLC ine

(kōwhiringa) snpe-dlc-quant –input_dlc model.dlc –float_fallback –override_params

3.3. (NUI) Ko te ONNX I/O kei te raupapa o te NCHW; Ko te DLC kua hurihia kei te raupapa NHWC

Tuhinga / Rauemi

Ko te Tuhituhi Taputapu Utauta Tino Aimet Qualcomm [pdf] Tohutohu quicksrnet_small_2x_w8a8, quicksrnet_small_4x_w8a8, quicksrnet_medium_2x_w8a8, quicksrnet_medium_4x_w8a8, Aimet Tuhituhi Taputapu Taputapu, Tuhituhi kete Utauta Pai, Tuhinga Taputapu

Tohutoro

• Pukapuka Kaiwhakamahi