edoakes

import ray
from ray import serve

ray.init(address='auto', namespace="serve-example", ignore_reinit_error=True)
serve.start(detached=True)

SentimentDeployment.deploy()

edoakes

from transformers import pipeline

@serve.deployment(route_prefix="/sentiment", name="sentiment")
class SentimentDeployment:
    def __init__(self):
        self.classifier = pipeline("sentiment-analysis")

    async def __call__(self, request):
        data = await request.body()
        [result] = self.classifier(str(data))

edoakes

import ray
from ray import serve
ray.init(address='auto', namespace="serve-example", ignore_reinit_error=True) # Connect to the local running Ray cluster.
serve.start(detached=True) # Start the Ray Serve processes within the Ray cluster.

edoakes

import requests
input_text = "Ray Serve eases the pain of model serving"
result = requests.get("http://127.0.0.1:8000/sentiment", data=input_text).text
print("Result for '{}': {}".format(input_text, result))

edoakes

import ray
from ray import serve

# Connect to the running Ray Serve instance.
ray.init(address='auto', namespace="serve-example", ignore_reinit_error=True)
serve.start(detached=True)

# Deploy the model.
SentimentDeployment.deploy()

edoakes

import joblib
import s3fs
import sklearn

@serve.deployment(route_prefix="/sentiment", name="sentiment-deployment")
class SentimentDeployment:
    def __init__(self):
        fs = s3fs.S3FileSystem(anon=True)
        with fs.open('ray-serve-blog/unigram_vectorizer.joblib', 'rb') as f:
            self.vectorizer = joblib.load(f)

edoakes

import ray
from ray import serve

# Connect to the running Ray Serve instance.
ray.init(address='auto', ignore_reinit_error=True)
serve.init()
# Deploy the model.
serve.create_backend("sklearn_backend", SKLearnBackend)
serve.create_endpoint("sentiment_endpoint", backend="sklearn_backend", route="/sentiment")

edoakes

import requests

from ray import serve

serve.init()

# Main concepts

## Endpoints

edoakes

import argparse
import time
import random
import math

parser = argparse.ArgumentParser(description="Approximate digits of Pi using Monte Carlo simulation.")
parser.add_argument("--num-samples", type=int, default=1000000)
parser.add_argument("--parallel", default=False, action="store_true")
parser.add_argument("--distributed", default=False, action="store_true")

edoakes

cluster_name: monte_carlo_pi

# The number of worker nodes to launch in addition to the head node.
min_workers: 9
max_workers: 9

provider:
 type: aws
 region: us-west-2
 availability_zone: us-west-2a