ibm / max-news-text-generator Goto Github PK

View Code? Open in Web Editor NEW

27.0 24.0 25.0 380 KB

Generate English-language text similar to the news articles in the One Billion Words data set.

Home Page: https://developer.ibm.com/code/exchanges/models/

License: Apache License 2.0

Python 94.43% Dockerfile 4.12% Starlark 1.45%

machine-learning docker-image machine-learning-models

max-news-text-generator's Introduction

IBM Code Model Asset Exchange: News Text Generator

This repository contains code to instantiate and deploy a text generation model. This model recognizes a text file as an input and outputs a string. The model was trained on the One Billion Word Benchmark (http://arxiv.org/abs/1312.3005) data set. The input to the model is a simple text file, and the output is a string containing the words that are predicted to follow. The model has a vocabulary of approximately 800,000 words.

The model files are hosted on IBM Cloud Object Storage. The code in this repository deploys the model as a web service in a Docker container. This repository was developed as part of the IBM Code Model Asset Exchange and the public API is powered by IBM Cloud.

Model Metadata

Domain	Application	Industry	Framework	Training Data	Input Data Format
Text	Text generation	Multi	TensorFlow	1 Billion Word Language Model Benchmark	text file

References

Rafal Jozefowicz, Oriol Vinyals, Mike Schuster, Noam Shazeer: “Exploring the Limits of Language Modeling”, 2016; arXiv:1602.02410.

Licenses

Component	License	Link
This repository	Apache 2.0	LICENSE
Pretrained weights	Apache 2.0	LICENSE
Training Data		1 Billion Word Language Model Benchmark

Pre-requisites:

Note: this model can be very memory intensive. If you experience crashes (such as the model API process terminating with a Killed message), ensure your docker container has sufficient resources allocated (for example you may need to increase the default memory limit on Mac or Windows).

docker: The Docker command-line interface. Follow the installation instructions for your system.
The minimum recommended resources for this model is 8 GB Memory and 4 CPUs.
If you are on x86-64/AMD64, your CPU must support AVX at the minimum.

Deployment options

Deploy from Quay
Deploy on Red Hat OpenShift
Deploy on Kubernetes
Run Locally

Deploy from Quay

To run the docker image, which automatically starts the model serving API, run:

$ docker run -it -p 5000:5000 quay.io/codait/max-news-text-generator

This will pull a pre-built image from the Quay.io container registry (or use an existing image if already cached locally) and run it. If you'd rather checkout and build the model locally you can follow the run locally steps below.

Deploy on Red Hat OpenShift

You can deploy the model-serving microservice on Red Hat OpenShift by following the instructions for the OpenShift web console or the OpenShift Container Platform CLI in this tutorial, specifying quay.io/codait/max-news-text-generator as the image name.

Note that this model requires at least 8GB of RAM. Therefore this model will not run in a cluster that was provisioned under the OpenShift Online starter plan, which is capped at 2GB.

Deploy on Kubernetes

You can also deploy the model on Kubernetes using the latest docker image on Quay.

On your Kubernetes cluster, run the following commands:

$ kubectl apply -f https://raw.githubusercontent.com/IBM/MAX-News-Text-Generator/master/max-news-text-generator.yaml

The model will be available internally at port 5000, but can also be accessed externally through the NodePort.

A more elaborate tutorial on how to deploy this MAX model to production on IBM Cloud can be found here.

1. Build the Model

Clone this repository locally. In a terminal, run the following command:

$ git clone https://github.com/IBM/MAX-News-Text-Generator.git

Change directory into the repository base folder:

$ cd MAX-News-Text-Generator

To build the docker image locally, run:

$ docker build -t max-news-text-generator .

All required model assets will be downloaded during the build process. Note that currently this docker image is CPU only (we will add support for GPU images later).

2. Deploy the Model

To run the docker image, which automatically starts the model serving API, run:

$ docker run -it -p 5000:5000 max-news-text-generator

3. Use the Model

The API server automatically generates an interactive Swagger documentation page. Go to http://localhost:5000 to load it. From there you can explore the API and also create test requests.

Use the model/predict endpoint to load some seed text (you can use one of the test files from the samples folder) and get predicted output from the API.

You can also test it on the command line, for example:

$ curl -F "text=@samples/sample1.txt" -XPOST http://localhost:5000/model/predict

You should see a JSON response like that below:

{"status": "ok", "pred_txt": "This is a test rather than an alternative view . </S> "}

4. Development

To run the Flask API app in debug mode, edit config.py to set DEBUG = True under the application settings. You will then need to rebuild the docker image (see step 1).

5. Cleanup

To stop the docker container type CTRL + C in your terminal.

Resources and Contributions

If you are interested in contributing to the Model Asset Exchange project or have any queries, please follow the instructions here.

max-news-text-generator's People

Contributors

Stargazers

Watchers

max-news-text-generator's Issues

Package/compress model artifacts on COS

The [checkpoint] assets are currently stored in individual files on COS, e.g.
http://max-assets.s3-api.us-geo.objectstorage.softlayer.net/lm_1b/ckpt-base
http://max-assets.s3-api.us-geo.objectstorage.softlayer.net/lm_1b/ckpt-char-embedding. The individual files should be packaged and compressed in a tar.gz archive, following the approach used by other models. Refer to https://github.com/IBM/MAX-Object-Detector/blob/master/Dockerfile for an example.

Benefits:

Only a single file needs to be downloaded (less work)
Artifacts are compressed, potentially reducing the time it takes to download them

/model/predict endpoint fails on Red Hat OpenShift due to a permission error


Traceback (most recent call last):
--
  | File "lm_1b/lm_1b_eval.py", line 309, in <module>
  | tf.app.run()
  | File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 125, in run
  | _sys.exit(main(argv))
  | File "lm_1b/lm_1b_eval.py", line 299, in main
  | _SampleModel(FLAGS.prefix, vocab)
  | File "lm_1b/lm_1b_eval.py", line 209, in _SampleModel
  | txt_file = open("out.txt", "w")
  | PermissionError: [Errno 13] Permission denied: 'out.txt'

/model/metadata endpoint returns no source information

{
  "id": "lm_1b",
  "name": "lm_1b TensorFlow Model",
  "description": "Generative language model trained on the One Billion Words data set",
  "type": "Generative Language Model",
  "source": null,
  "license": "Apache v2"
}

Add request param descriptions and update Swagger models help text

Example in PR:
IBM/MAX-Object-Detector#20

Error when you don't have enough memory is confusing

As discovered in #5 the following error is thrown after a prediction is Killed due to running out of memory. The fact that this can happen is documented in the README, but it may be worth adding a troubleshooting section that includes what the actual error looks like for reference.

[2018-07-28 03:07:36,563] ERROR in app: Exception on /model/predict [POST]
Traceback (most recent call last):
  File "/opt/conda/lib/python3.6/site-packages/flask/app.py", line 1813, in full_dispatch_request
    rv = self.dispatch_request()
  File "/opt/conda/lib/python3.6/site-packages/flask/app.py", line 1799, in dispatch_request
    return self.view_functions[rule.endpoint](**req.view_args)
  File "/opt/conda/lib/python3.6/site-packages/flask_restplus/api.py", line 319, in wrapper
    resp = resource(*args, **kwargs)
  File "/opt/conda/lib/python3.6/site-packages/flask/views.py", line 88, in view
    return self.dispatch_request(*args, **kwargs)
  File "/opt/conda/lib/python3.6/site-packages/flask_restplus/resource.py", line 44, in dispatch_request
    resp = meth(*args, **kwargs)
  File "/opt/conda/lib/python3.6/site-packages/flask_restplus/marshalling.py", line 136, in wrapper
    resp = f(*args, **kwargs)
  File "/workspace/api/model.py", line 58, in post
    preds = self.model_wrapper.predict(text)
  File "/workspace/core/tensorflow.py", line 29, in predict
    txt_file = open("out.txt", "r")
FileNotFoundError: [Errno 2] No such file or directory: 'out.txt'

pip install from requirements.txt

Example PR:
IBM/MAX-Review-Text-Generator#7

Get Travis CI working

Travis CI is unable to test this model due to memory constrains. See discussion on PR #11

Missing link to model files

The README.md states the following:

The model files are hosted on IBM Cloud Object Storage

However, there is no link provided to the model files in IBM Cloud Object Storage.
Some of the other IBM MAX models do contain direct links to download the model
files but I do see the link in this model.

When the input is sufficiently long, output simply repeats part of the input

When the input is sufficiently long, the output simply repeats the input. In other words, there is no way to specify output length.

Example input: "This repository contains code to instantiate and deploy a text generation model. This model recognizes a text file as an input and outputs a string."

outputs "This repository contains code to instantiate and deploy a text generation model . This model recognizes"

Change input/output data model for predict requests

I noticed while looking at #14 that the response format is a bit odd for this model: it returns a list of {"pred_text":...} objects. But there is only one predicted text string that will be returned.

I suggest changing the output format to something similar to https://github.com/IBM/MAX-Review-Text-Generator. While doing that we could also update the input style to take a JSON object {"seed_text":...} like the review generator.