Merge pull request #4 from linto-ai/next

merge next -> master

.env_default

@@ -0,0 +1,15 @@
+# SERVING PARAMETERS
+SERVICE_MODE=task
+
+# SERVICE PARAMETERS
+SERVICES_BROKER=redis://172.17.0.1:6379
+BROKER_PASS=
+
+# SERVICE DISCOVERY
+SERVICE_NAME=linto-punctuation
+LANGUAGE=fr-FR
+# QUEUE_NAME=(Optionnal)
+# MODEL_INFO=This model does something
+
+# CONCURRENCY
+CONCURRENCY=2

Dockerfile

@@ -1,15 +1,13 @@
-FROM python:3.8
-LABEL maintainer="[email protected]"
-ENV PYTHONUNBUFFERED TRUE
-ENV IMAGE_NAME linto-platform-diarization
+FROM python:3.9
+LABEL maintainer="[email protected]"
 
-RUN apt-get update \
-    && apt-get install --no-install-recommends -y \
-    ca-certificates \
-    g++ \
-    openjdk-11-jre-headless \
-    curl \
-    wget
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        ca-certificates \
+        g++ \
+        curl \
+        libtinfo5 \
+        wget
 
 # Rust compiler for tokenizers
 RUN curl https://sh.rustup.rs -sSf | bash -s -- -y
@@ -18,8 +16,8 @@ ENV PATH="/root/.cargo/bin:${PATH}"
 WORKDIR /usr/src/app
 
 # Python dependencies
-COPY requirements.txt ./
-RUN pip install --no-cache-dir -r requirements.txt
+COPY requirements.txt .
+RUN pip3 install --no-cache-dir -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html
 
 # Supervisor
 COPY celery_app /usr/src/app/celery_app
@@ -28,13 +26,8 @@ COPY document /usr/src/app/document
 COPY punctuation /usr/src/app/punctuation
 RUN mkdir /usr/src/app/model-store
 RUN mkdir -p /usr/src/app/tmp
-COPY config.properties /usr/src/app/config.properties
-COPY RELEASE.md ./
 COPY docker-entrypoint.sh wait-for-it.sh healthcheck.sh ./
 
-# Grep CURRENT VERSION
-RUN export VERSION=$(awk -v RS='' '/#/ {print; exit}' RELEASE.md | head -1 | sed 's/#//' | sed 's/ //')
-
 ENV PYTHONPATH="${PYTHONPATH}:/usr/src/app/punctuation"
 HEALTHCHECK CMD ./healthcheck.sh
 

Dockerfile.cpu

@@ -0,0 +1,35 @@
+FROM python:3.9
+LABEL maintainer="[email protected]"
+
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        ca-certificates \
+        g++ \
+        openjdk-11-jre-headless \
+        curl \
+        wget
+
+# Rust compiler for tokenizers
+RUN curl https://sh.rustup.rs -sSf | bash -s -- -y
+ENV PATH="/root/.cargo/bin:${PATH}"
+
+WORKDIR /usr/src/app
+
+# Python dependencies
+COPY requirements.cpu.txt .
+RUN pip3 install --no-cache-dir -r requirements.cpu.txt -f https://download.pytorch.org/whl/torch_stable.html
+
+# Supervisor
+COPY celery_app /usr/src/app/celery_app
+COPY http_server /usr/src/app/http_server
+COPY document /usr/src/app/document
+COPY punctuation /usr/src/app/punctuation
+RUN mkdir /usr/src/app/model-store
+RUN mkdir -p /usr/src/app/tmp
+COPY docker-entrypoint.sh wait-for-it.sh healthcheck.sh ./
+
+ENV PYTHONPATH="${PYTHONPATH}:/usr/src/app/punctuation"
+HEALTHCHECK CMD ./healthcheck.sh
+
+ENV TEMP=/usr/src/app/tmp
+ENTRYPOINT ["./docker-entrypoint.sh"]

Jenkinsfile

@@ -47,5 +47,25 @@ pipeline {
                 }
             }
         }
+
+        stage('Docker build for recasepunc branch'){
+            when{
+                branch 'recasepunc'
+            }
+            steps {
+                echo 'Publishing recasepunc'
+                script {
+                    image = docker.build(env.DOCKER_HUB_REPO, "-f Dockerfile .")
+                    VERSION = sh(
+                        returnStdout: true,
+                        script: "awk -v RS='' '/#/ {print; exit}' RELEASE.md | head -1 | sed 's/#//' | sed 's/ //'"
+                    ).trim()
+                    docker.withRegistry('https://registry.hub.docker.com', env.DOCKER_HUB_CRED) {
+                        image.push('recasepunc-latest')
+                    }
+                }
+            }
+        }
+
     }// end stages
 }

README.md

@@ -22,13 +22,27 @@ LinTO-platform-punctuation can either be used as a standalone punctuation servic
 ## Pre-requisites
 
 ### Models
-The punctuation service relies on a trained punctuation prediction model.
+The punctuation service relies on a trained recasing and punctuation prediction model.
 
-We provide homebrew models on [dl.linto.ai](https://dl.linto.ai/downloads/model-distribution/punctuation_models/).
+Some models trained on [Common Crawl](http://data.statmt.org/cc-100/) are available on [recasepunc](https://github.com/benob/recasepunc) for the following the languages:
+* French
+  * [fr-txt.large.19000](https://github.com/benob/recasepunc/releases/download/0.3/fr-txt.large.19000)
+  * [fr.22000](https://github.com/benob/recasepunc/releases/download/0.3/fr.22000)
+* English
+  * [en.23000](https://github.com/benob/recasepunc/releases/download/0.3/en.23000)
+* Italian
+  * [it.22000](https://github.com/CoffeePerry/recasepunc/releases/download/v0.1.0/it.22000)
+* Chinese
+  * [zh.24000](https://github.com/benob/recasepunc/releases/download/0.3/zh.24000)
+
+<!-- We provide homebrew models on [dl.linto.ai](https://dl.linto.ai/downloads/model-distribution/punctuation_models/). -->
 
 ### Docker
 The punctuation service requires docker up and running.
 
+For GPU capabilities, it is also needed to install
+[nvidia-container-toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html).
+
 ### (micro-service) Service broker
 The punctuation only entry point in job mode are tasks posted on a REDIS message broker using [Celery](https://github.com/celery/celery). 
 
@@ -52,13 +66,13 @@ docker pull registry.linto.ai/lintoai/linto-platform-punctuation:latest
 
 **2- Download the models**
 
-Have the punctuation model (.mar) ready at MODEL_PATH.
+Have the punctuation model ready at `<MODEL_PATH>`.
 
 ### HTTP
 
 **1- Fill the .env**
 ```bash
-cp .env_default_http .env
+cp .env_default .env
 ```
 
 Fill the .env with your values.
@@ -73,12 +87,14 @@ Fill the .env with your values.
 
 ```bash
 docker run --rm \
--v MODEL_PATH:/usr/src/app/model-store/punctuation.mar \
+-v <MODEL_PATH>:/usr/src/app/model-store/model \
 -p HOST_SERVING_PORT:80 \
 --env-file .env \
 linto-platform-punctuation:latest
 ```
 
+Also add ```--gpus all``` as an option to enable GPU capabilities.
+
 This will run a container providing an http API binded on the host HOST_SERVING_PORT port.
 
 
@@ -90,7 +106,7 @@ You need a message broker up and running at MY_SERVICE_BROKER. Instance are typi
 
 **1- Fill the .env**
 ```bash
-cp .env_default_task .env
+cp .env_default .env
 ```
 
 Fill the .env with your values.
@@ -118,7 +134,7 @@ services:
   punctuation-service:
     image: linto-platform-punctuation:latest
     volumes:
-      - /my/path/to/models/punctuation.mar:/usr/src/app/model-store/punctuation.mar
+      - /my/path/to/models/punctuation.mar:/usr/src/app/model-store/model
     env_file: .env
     deploy:
       replicas: 1
@@ -156,7 +172,7 @@ The following information are registered:
   "service_language": $LANGUAGE,
   "queue_name": $QUEUE_NAME,
   "version": "1.2.0", # This repository's version
-  "info": "Bert Based Punctuation model for french punctuation prediction",
+  "info": "Punctuation model for french punctuation prediction",
   "last_alive": 65478213,
   "concurrency": 1
 }
@@ -223,3 +239,6 @@ curl -X POST "http://YOUR_SERVICE:YOUR_PORT/punctuation" -H  "accept: applicatio
 
 ## License
 This project is developped under the AGPLv3 License (see LICENSE).
+
+## Acknowledgments
+* [recasepunc](https://github.com/benob/recasepunc) Python library to train recasing and punctuation models, and to apply them (License BSD 3).

RELEASE.md

@@ -1,3 +1,6 @@
+# 2.0.0
+- Integration of recasepunc
+
 # 1.1.1
 - Fix error on empty sentences
 

celery_app/register.py

@@ -26,6 +26,7 @@ def register(is_heartbeat: bool = False) -> bool:
     """
     host, port = os.environ.get("SERVICES_BROKER").split("//")[1].split(":")
     password = os.environ.get("BROKER_PASS", None)
+    if not password: password = None
     r = redis.Redis(
         host=host, port=int(port), db=SERVICE_DISCOVERY_DB, password=password
     )
@@ -59,8 +60,10 @@ def unregister() -> None:
     """Un-register the service"""
     try:
         host, port = os.environ.get("SERVICES_BROKER").split("//")[1].split(":")
+        password = os.environ.get("BROKER_PASS", None)
+        if not password: password = None
         r = redis.Redis(
-            host=host, port=int(port), db=SERVICE_DISCOVERY_DB, password="password"
+            host=host, port=int(port), db=SERVICE_DISCOVERY_DB, password=password
         )
         r.json().delete(f"service:{host_name}")
     except Exception as error:

celery_app/tasks.py

@@ -1,53 +1,26 @@
-import json
 from typing import Union
 
-import requests
-
 from celery_app.celeryapp import celery
 
+from punctuation.recasepunc import load_model, generate_predictions
+
+MODEL = load_model()
 
 @celery.task(name="punctuation_task", bind=True)
 def punctuation_task(self, text: Union[str, list]):
     """punctuation_task do a synchronous call to the punctuation serving API"""
     self.update_state(state="STARTED")
-    # Fetch model name
-    try:
-        result = requests.get(
-            "http://localhost:8081/models",
-            headers={
-                "accept": "application/json",
-            },
-        )
-        models = json.loads(result.text)
-        model_name = models["models"][0]["modelName"]
-    except Exception as error:
-        raise Exception("Failed to fetch model name") from error
+
+    unique = isinstance(text, str)
 
-    if isinstance(text, str):
+    if unique:
         sentences = [text]
     else:
         sentences = text
-    punctuated_sentences = []
-    for i, sentence in enumerate(sentences):
-        self.update_state(state="STARTED", meta={"current": i, "total": len(sentences)})
-
-        result = requests.post(
-            f"http://localhost:8080/predictions/{model_name}",
-            headers={"content-type": "application/octet-stream"},
-            data=sentence.strip().encode("utf-8"),
-        )
-        if result.status_code == 200:
-            punctuated_sentence = result.text
-        else:
-            print("Failed to predict punctuation on sentence: >{sentence}<")
-            punctuated_sentence = sentence
-        # First letter in capital
-        if len(punctuated_sentence):
-            punctuated_sentence = punctuated_sentence[0].upper() + punctuated_sentence[1:]
-        punctuated_sentences.append(punctuated_sentence)
+    punctuated_sentences = generate_predictions(MODEL, sentences)
 
     return (
         punctuated_sentences[0]
-        if len(punctuated_sentences) == 1
+        if unique
         else punctuated_sentences
     )

docker-compose.yml

@@ -4,7 +4,7 @@ services:
   punctuation-service:
     image: linto-platform-punctuation:latest
     volumes:
-      - /path/to/your/model.mar/usr/src/app/model-store/punctuation.mar
+      - /path/to/your/model.mar/usr/src/app/model-store/model
     env_file: .env
     deploy:
       replicas: 1

docker-entrypoint.sh

@@ -2,15 +2,11 @@
 
 echo "RUNNING service"
 
-export VERSION=$(awk -v RS='' '/#/ {print; exit}' RELEASE.md | head -1 | sed 's/#//' | sed 's/ //')
-
 if [ -z "$SERVICE_MODE" ]
 then
     echo "ERROR: Must specify a serving mode: [ http | task ]"
     exit -1
 else
-    # Model serving
-    torchserve --start --ncs --ts-config /usr/src/app/config.properties
     if [ "$SERVICE_MODE" = "http" ]
     then
         echo "Running http server"
@@ -19,25 +15,24 @@ else
     elif [ "$SERVICE_MODE" == "task" ]
     then
         echo "Running celery worker" 
-        /usr/src/app/wait-for-it.sh $(echo $SERVICES_BROKER | cut -d'/' -f 3) --timeout=20 --strict -- echo " $SERVICES_BROKER (Service Broker) is up"
+        /usr/src/app/wait-for-it.sh $(echo $SERVICES_BROKER | cut -d'/' -f 3) --timeout=20 --strict -- echo " $SERVICES_BROKER (Service Broker) is up" || exit $?
         # MICRO SERVICE
         ## QUEUE NAME
         QUEUE=$(python -c "from celery_app.register import queue; exit(queue())" 2>&1)
         echo "Service set to $QUEUE"
 
         ## REGISTRATION
-        python -c "from celery_app.register import register; register()"
+        python -c "from celery_app.register import register; register()" # || exit $?
         echo "Service registered"
 
         ## WORKER
-        celery --app=celery_app.celeryapp worker -n punctuation_$SERVICE_NAME@%h --queues=$QUEUE -c $CONCURRENCY
+        celery --app=celery_app.celeryapp worker --pool=solo -n punctuation_$SERVICE_NAME@%h --queues=$QUEUE -c $CONCURRENCY
 
         ## UNREGISTERING
-        python -c "from celery_app.register import unregister; unregister()"
+        python -c "from celery_app.register import unregister; unregister()" || exit $?
         echo "Service unregistered"
     else
         echo "ERROR: Wrong serving command: $SERVICE_MODE"
         exit -1
     fi
-    torchserve --stop
 fi