Skip to content

Latest commit

 

History

History
84 lines (54 loc) · 4.07 KB

README.md

File metadata and controls

84 lines (54 loc) · 4.07 KB
Raw

Text Classification using TorchServe's default text_classifier handler

!!!Deprecation Warning!!!

This example requires TorchText which is deprecated. Please use version <= 0.11.1 of TorchServe for this example

This is an example to create a text classification dataset and train a sentiment model. We have used the following torchtext example to train the model.

https://github.com/pytorch/text/tree/master/examples/text_classification

We have copied the files from above example and made small changes to save the model's state dict and added default values.

Training the model

Run the following commands to train the model :

python run_script.py

The above command generated the model's state dict as model.pt and the vocab used during model training as source_vocab.pt

Serve the text classification model on TorchServe

  • Create a torch model archive using the torch-model-archiver utility to archive the above files.

    torch-model-archiver --model-name my_text_classifier --version 1.0 --model-file model.py --serialized-file model.pt  --handler text_classifier --extra-files "index_to_name.json,source_vocab.pt"

    NOTE - run_script.sh has generated source_vocab.pt and it is a mandatory file for this handler. If you are planning to override or use custom source vocab. then name it as source_vocab.pt and provide it as --extra-files as per above example. Other option is to extend TextHandler and override get_source_vocab_path function in your custom handler. Refer custom handler for detail

  • Register the model on TorchServe using the above model archive file and run digit recognition inference

    mkdir model_store
mv my_text_classifier.mar model_store/
torchserve --start --model-store model_store --models my_tc=my_text_classifier.mar --disable-token-auth  --enable-model-api
curl http://127.0.0.1:8080/predictions/my_tc -T examples/text_classification/sample_text.txt

To make a captum explanations request on the Torchserve side, use the below command:

curl -X POST http://127.0.0.1:8080/explanations/my_tc -T examples/text_classification/sample_text.txt

In order to run Captum Explanations with the request input in a json file, follow the below steps:

In the config.properties, specify service_envelope=body and make the curl request as below:

curl -H "Content-Type: application/json" --data @examples/text_classification/text_classifier_ts.json http://127.0.0.1:8080/explanations/my_tc_explain

When a json file is passed as a request format to the curl, Torchserve unwraps the json file from the request body. This is the reason for specifying service_envelope=body in the config.properties file

Captum Explanations

The explain is called with the following request api http://127.0.0.1:8080/explanations/my_tc_explain

Torchserve supports Captum Explanations for Eager models only.

Captum/Explain doesn't support batching.

The handler changes:

  1. The handlers should initialize.
self.lig = LayerIntegratedGradients(captum_sequence_forward, self.model.bert.embeddings)

in the initialize function for the captum to work.

  1. The Base handler handle uses the explain_handle method to perform captum insights based on whether user wants predictions or explanations. These methods can be overriden to make your changes in the handler.

  2. The get_insights method in the handler is called by the explain_handle method to calculate insights using captum.

  3. If the custom handler overrides handle function of base handler, the explain_handle function should be called to get captum insights.

NOTE: The current default model for text classification uses EmbeddingBag which Computes sums or means of ‘bags’ of embeddings, without instantiating the intermediate embedding, so it returns the captum explanations on a sentence embedding level and not on a word embedding level.

Running KServe

Refer the End to End KServe document to run it in the cluster.