Merge pull request #236 from isi-nlp/130-restapi

Add REST API around decoder

Dockerfile

@@ -0,0 +1,11 @@
+ARG experiment_dir
+FROM python:3.7
+# RUN echo ${experiment_dir}
+# RUN echo "hello world"
+# COPY ${experiment_dir} /experiment/
+COPY ./experiments/sample-exp/ /experiment/
+COPY . /rtg
+WORKDIR /rtg
+RUN pip install -e ./
+RUN python -m rtg.pipeline /experiment/
+CMD python -m rtg.deploy /experiment/

docs/clitools.adoc

@@ -16,9 +16,10 @@ The following command line tools are added when `rtg` is installed using pip.
 | Command        | Purpose
 | rtg-pipe       | Run rtg-prep, rtg-train and test case evaluation
 | rtg-decode     | Decode new source files using the values set in `conf.yml`
-| rtg-decode-pro | Decode new source files using the values that you supply from CLI args
 | rtg-export     | Export an experiment
 | rtg-fork       | Fork an experiment with/without same conf, code, data, vocabularies etc
+| rtg-serve      | Serve an RTG model over HTTP API using  Flask server
+| rtg-decode-pro | Decode new source files using the values that you supply from CLI args
 | rtg-prep       | Prepare an experiment. You should be using `rtg-pipe`
 | rtg-train      | Train a model. You should be using `rtg-pipe`
 | rtg-syscomb    | System combination. Dont bother about it for now.

docs/index.adoc

@@ -17,5 +17,7 @@ include::clitools.adoc[]
 
 include::environ.adoc[]
 
+include::serve.adoc[]
+
 include::develop.adoc[]
 

docs/serve.adoc

@@ -0,0 +1,66 @@
+
+== RTG Serve
+
+RTG model can be served using Flask Server.
+
+
+[source,commandline]
+----
+$ python -m rtg.serve -h  # rtg-serve
+[07-13 22:38:01] p49095 {__init__:53} INFO - rtg v0.3.1 from /Users/tg/work/me/rtg
+usage: rtg.serve [-h] [-sc] [-p PORT] [-ho HOST] [-msl MAX_SRC_LEN] exp_dir
+
+deploy a model to a RESTful react server
+
+positional arguments:
+  exp_dir               Experiment directory
+
+optional arguments:
+  -h, --help            show this help message and exit
+  -sc, --skip-check     Skip Checking whether the experiment dir is prepared
+                        and trained (default: False)
+  -p PORT, --port PORT  port to run server on (default: 6060)
+  -ho HOST, --host HOST
+                        Host address to bind. (default: 0.0.0.0)
+  -msl MAX_SRC_LEN, --max-src-len MAX_SRC_LEN
+                        max source len; longer seqs will be truncated
+                        (default: None)
+----
+
+
+To launch a service for `runs/001-tfm` experiment
+
+`python -m rtg.serve -sc runs/001-tfm`
+
+It prints :
+`* Running on http://0.0.0.0:6060/ (Press CTRL+C to quit)`
+
+Currently only `/translate` API is supported. It accepts both `GET` with query params and `POST` with form params.
+
+NOTE: batch decoding is yet to be supported. Current decoder decodes only one sentence at a time.
+
+An example POST request:
+----
+ curl --data "source=Comment allez-vous?" --data "source=Bonne journée" http://localhost:6060/translate
+
+----
+[source,json]
+----
+
+{
+  "source": [
+    "Comment allez-vous?",
+    "Bonne journée"
+  ],
+  "translation": [
+    "How are you?",
+    "Have a nice day"
+  ]
+}
+----
+You can also request like GET method as `http://localhost:6060/translate?source=text1&source=text2`
+after properly URL encoding the `text1` `text2`. This should only be used for quick testing in your web browser.
+
+
+
+

rtg/pipeline.py

@@ -285,7 +285,7 @@ def run(self, run_tests=True):
 
 
 def parse_args():
-    parser = argparse.ArgumentParser(prog="rtg.prep", description="prepare NMT experiment")
+    parser = argparse.ArgumentParser(prog="rtg-pipe", description="RTG Pipeline CLI")
     parser.add_argument("exp", metavar='EXP_DIR', help="Working directory of experiment", type=Path)
     parser.add_argument("conf", metavar='conf.yml', type=Path, nargs='?',
                         help="Config File. By default <work_dir>/conf.yml is used")

rtg/serve.py

@@ -0,0 +1,82 @@
+#!/usr/bin/env python
+"""
+Serves an RTG model using Flask HTTP server
+"""
+from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter
+
+from flask import Flask, request, jsonify
+import torch
+
+from rtg import TranslationExperiment as Experiment, log
+from rtg.module.decoder import Decoder
+
+
+def prepare_decoder(cli_args):
+    # No grads required for decode
+    torch.set_grad_enabled(False)
+    exp = Experiment(cli_args.pop("exp_dir"), read_only=True)
+    dec_args = exp.config.get("decoder") or exp.config["tester"].get("decoder", {})
+    validate_args(cli_args, dec_args, exp)
+    decoder = Decoder.new(exp, ensemble=dec_args.pop("ensemble", 1))
+    return decoder, dec_args
+
+
+def attach_translate_route(app, decoder, dec_args):
+
+    app.config['JSON_AS_ASCII'] = False
+
+    @app.route("/translate", methods=["POST", "GET"])
+    def translate():
+        if request.method not in ("POST", "GET"):
+            return "GET and POST are supported", 400
+        if request.method == 'GET':
+            sources = request.args.getlist("source", None)
+        else:
+            sources = request.form.getlist("source", None)
+        if not sources:
+            return "Please provide parameter 'source'", 400
+
+        translations = []
+        for source in sources:
+            translated = decoder.decode_sentence(source, **dec_args)[0][1]
+            translations.append(translated)
+        res = dict(source=sources, translation=translations)
+        return jsonify(res)
+
+
+def validate_args(cli_args, conf_args, exp: Experiment):
+    if not cli_args.pop("skip_check"):  # if --skip-check is not requested
+        assert exp.has_prepared(), (f'Experiment dir {exp.work_dir} is not ready to train.'
+                                    f' Please run "prep" sub task')
+        assert exp.has_trained(), (f"Experiment dir {exp.work_dir} is not ready to decode."
+                                   f" Please run 'train' sub task or --skip-check to ignore this")
+
+def parse_args():
+    parser = ArgumentParser(
+        prog="rtg.serve",
+        description="deploy a model to a RESTful react server",
+        formatter_class=ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument("exp_dir", help="Experiment directory", type=str)
+    parser.add_argument("-sc", "--skip-check", action="store_true",
+        help="Skip Checking whether the experiment dir is prepared and trained")
+    parser.add_argument("-p", "--port", type=int, help="port to run server on", default=6060)
+    parser.add_argument("-ho", "--host", help="Host address to bind.", default='0.0.0.0')
+    parser.add_argument("-msl", "--max-src-len", type=int,
+                        help="max source len; longer seqs will be truncated")
+    args = vars(parser.parse_args())
+    return args
+
+
+def main():
+    cli_args = parse_args()
+    decoder, dec_args = prepare_decoder(cli_args)
+    app = Flask(__name__)
+    #CORS(app)  # TODO: insecure
+    app.debug = True
+    attach_translate_route(app, decoder, dec_args)
+    app.run(port=cli_args["port"], host=cli_args["host"])
+
+
+if __name__ == "__main__":
+    main()

setup.py

@@ -44,6 +44,9 @@
         'nlcodec >= 0.2.2',
         'torch >= 1.4'
     ],
+    extra_requires={
+        'serve': ['flask >= 1.1.2'],
+    },
     python_requires='>=3.7',
     entry_points={
         'console_scripts': [
@@ -54,6 +57,7 @@
             'rtg-prep=rtg.prep:main',
             'rtg-train=rtg.train:main',
             'rtg-fork=rtg.fork:main',
+            'rtg-serve=rtg.serve:main',
             'rtg-syscomb=rtg.syscomb.__main__:main',
         ],
     }