Storing the Downloaded Models

[selected-pixel-jameson]

Is there a way I can store the ResNet50 image that is download somewhere so that it's not re-downloaded when I initialize img2Vec via img2vec = Img2Vec(cuda=False, model="resnet50", layer_output_size=2048)

I want to create an Docker image that uses this library which will run through AWS lambda and I'd like to store the Model so that every time the lambda function kicks-off it's not re downloading the model when img2vec = Img2Vec(cuda=False, model="resnet50", layer_output_size=2048) is called.

[christiansafka]

The model is probably small enough to package into your docker image and still run on Lambda. If it isn't, you can look into lambda layers.

To package it into your image you'll have to figure out where in your container pytorch is downloading the model to, and make sure you put the model weights there when building the image (use dockerfile).

[selected-pixel-jameson]

Thanks for the response.

Do you by chance have a reference to model weights? And how I would potentially go about finding them?

Really appreciate your responses on all my questions.

[christiansafka]

Sure, so according to PyTorch docs: "Instancing a pre-trained model will download its weights to a cache directory. This directory can be set using the TORCH_HOME environment variable."

How I found this was by googling "pytorch resnet50 where are pretrained weights saved"

[selected-pixel-jameson]

Sorry for the slow response. This helped and I was able to store these in the docker image and run my application using an AWS lambda function.