Well the question here is easy. Should we put it as before in model.quantize and model.trace? or we should move all the previous and newly added features to InferenceOptimizer's trace and quantize. I would prefer the second choice so that in future we could Use patch_nano to handle (nearly) all the...

Description Need more detailed instructions or environment specifications for the following Orca examples: Data ray-dataset-xgboost Environment requirements pip install xgboost-ray pip install ray==1.9.2 Data preparation Sign in and download incd.csv from here. RayOnSpark parameter_server Envir...

We only have a how-to-guide now. We can put the full notebook to Examples as well.

Description YOLOX evaluate the model with COCOAPI. The evaluation-related code like this: for cur_iter, (imgs, _, info_imgs, ids) in enumerate( progress_bar(self.dataloader) ): with torch.no_grad(): imgs = imgs.type(tensor_type) # skip...

Background ONNX Simplifier is presented to simplify the ONNX model. It infers the whole computation graph and then replaces the redundant operators with their constant outputs (a.k.a. constant folding). Some projects like YOLOv5 have adpoted ONNX Simplifier as a supplement to ONNX Runtime. We are co...

Since nano has now be quite friendly to 3.8 and native windows, I think chronos can move on to support these 2 platforms.

Openvino has released a new version 2020.2.0, click here Initial tests show that installing both bigdl-chronos and the latest version of openvino does not cause the installation to fail due to incompatible versions of numpy, but we still need more tests. Test list: setup: Rebuild wheel after adding...

Introduction Inspired by scikit-learn-intelex, we explore the possibility to use a patching API for bigdl-nano to easy the usage of current functions and enable more functions(e.g. GPU->CPU tool). Here is a simple usage demo from bigdl.nano import patch_something # patch many modules(e.g.torch.T...

API doc is missing for bigdl.nano.tf.keras.layers.Embedding and bigdl.nano.tf.optimizers.SparseAdam

Background Built-in dataset can be downloaded and preprocessed by get_public_dataset, but considering that users need dataset to benchmark, we may need to add a new API to help users generate dataset even with poor Internet connection. API design def gen_synthetic_data(len=10000, amplitude=10.0, ang...