Instruksi Dokumentasi Toolkit Efisiensi Qualcomm Aimet

KBA-231226181840

1. Setup Envirnment

1.1. Instal Driver Nvidia lan CUDA

1.2. Instal Pustaka Python sing Gegandhengan

python3 -m pip install –upgrade –ignore-installed pip
python3 -m pip nginstal -ignore-instal gdown
python3 -m pip nginstal -ignore-instal opencv-python
python3 -m pip nginstal –abaikan-dipasang obor==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 nginstal -ignore-instal jax
python3 -m pip nginstal -ignore-instal ftfy
python3 -m pip nginstal -ignore-installed torchinfo
python3 -m pip nginstal -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 nginstal -ignore-instal 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 install –ignore-installed numpy == 1.21.6
python3 -m pip nginstal -ignore-instal psutil

1.3. Klon aimet-model-zoo

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

1.4. Ngundhuh Set14

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

1.5. Ngowahi baris 39 aimet-model-zoo/aimet_zoo_torch/quicksrnet/dataloader/utils.py

owah-owahan
kanggo img_path ing glob.glob(os.path.join(test_images_dir, “*”)):
kanggo
kanggo img_path ing glob.glob(os.path.join(test_images_dir, “*_HR.*”)):

1.6. Run evaluasi.

# mbukak ing YOURPATH/aimet-model-run
# Kanggo quicksrnet_small_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_small_2x_w8a8 \
–dataset-path ../Set14/image_SRF_4

# Kanggo quicksrnet_small_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_small_4x_w8a8 \
–dataset-path ../Set14/image_SRF_4

# Kanggo quicksrnet_medium_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_medium_2x_w8a8 \
–dataset-path ../Set14/image_SRF_4

# Kanggo quicksrnet_medium_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-config quicksrnet_medium_4x_w8a8 \
–dataset-path ../Set14/image_SRF_4

umpamane sampeyan bakal entuk nilai PSNR kanggo model simulasi. Sampeyan bisa ngganti model-config kanggo macem-macem ukuran QuickSRNet, pilihan punika underaimet-modelzoo/aimet_zoo_torch/quicksrnet/model/model_cards/.

2 Tambah Patch

2.1. Bukak "Ekspor menyang ONNX Steps REVISED.docx"

2.2. Skip git commit id

2.3. Kode Bagian 1

Tambah kabeh 1. kode ing baris pungkasan (sawise baris 366) aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/models.py

2.4. Bagian 2 lan 3 Kode

Tambah kabeh kode 2, 3 ing baris 93 aimet-model-zoo/aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py

2.5. Parameter Kunci ing Function load_model

model = load_model(MODEL_PATH_INT8,

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

MODEL_PATH_INT8 = aimet_zoo_torch/quicksrnet/model/bobot/quicksrnet_small_2x_w8a8/pre_opt_weights
MODEL_NAME = QuickSRNetSmall
MODEL_ARGS.get(MODEL_NAME).get(MODEL_CONFIG) = {'faktor_skala': 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

Mangga ngganti variabel kanggo ukuran QuickSRNet beda

2.6 Modifikasi Ukuran Model

1. "input_shape" ing aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/model_cards/*.json
2. Ing njero fungsi load_model(...) ing aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/inference.py
3. Parameter ing fungsi export_to_onnx(…, input_height, input_width) saka "Ekspor menyang ONNX Steps REVISED.docx"

2.7 Re-Run 1.6 maneh kanggo ngekspor model ONNX

3. Ngonversi ing SNPE

3.1. Ngonversi

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

3.2. (Opsional) Ekstrak mung DLC ​​kuantitatif

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

3.3. (PENTING) ONNX I/O ing urutan NCHW; DLC sing diowahi dadi NHWC

Dokumen / Sumber Daya

Dokumentasi Toolkit Efisiensi Qualcomm Aimet [pdf] Pandhuan quicksrnet_small_2x_w8a8, quicksrnet_small_4x_w8a8, quicksrnet_medium_2x_w8a8, quicksrnet_medium_4x_w8a8, Dokumentasi Toolkit Efisiensi Aimet, Dokumentasi Toolkit Efisiensi, Dokumentasi Toolkit, Dokumen

Referensi

• Manual pangguna