- model: Code for training and distilling the model
- clipper-asymcc: A prediction serving system that sits between user-facing applications and machine learning models and frameworks.It is built atop Clipper
The deployed model is based on i-RevNet. To train a deployed model, just use the following commands. Make sure that you have installed PyTorch and CUDA in advance.
cd model/src/train
python CIFAR_main.py --nBlocks 18 18 18 --nStrides 1 2 2 --nChannels 16 64 256The knowledge distillation is based on RepDistiller. To distill a decoder, just use the following commands.
cd model/src/distill
python train_decoder.py --path_t ./save/models/iRevNet18/speech_irevnet_t.t7 \
--dataset cifar10 \
--distill kd \
--model_s iRevNet16x64 \
--learning_rate 0.01 \
-r 0.1 \
-a 0.9 \
-b 0 \
--trial 1 \
--irev \
--epochs 210 \
--lr_decay_epochs 120,150,180 \
--ec_k 2 \
--batch_size 128 \
--etFirst, you should have several nodes, each of which is recommended to be equipped with at least a GPU and is enabled with CUDA>=11.4.
Users can run setup.sh script on each node to satisfy Clipper's software requirements.
cd clipper-asymcc/run
bash scripts/setup.shUsers can run clipper_deploy.sh on each node to start clipper cluster
python clipper_deploy.pyYou can set the configuration parameters in config/*.json at first. Then run the following commands to run experiments.
python run_exp.py --conf /path/to/your/config --path /path/to/your/file
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent