caleb-kaiser

from encoder import get_encoder
encoder = get_encoder()


def pre_inference(sample, signature, metadata):
    context = encoder.encode(sample["text"])
    return {"context": [context]}


def post_inference(prediction, signature, metadata):

caleb-kaiser

- kind: deployment
  name: text

- kind: api
  name: generator
  tensorflow:
    model: s3://cortex-test-project/124M/124M/
    request_handler: handler.py

caleb-kaiser

import tensorflow as tf
import os
import json
import regex as re
from functools import lru_cache
import requests
import boto3


@lru_cache()

caleb-kaiser

# predictor.py
from transformers import GPT2Tokenizer, GPT2LMHeadModel

# download the pretrained DistilGPT2 model and set it to evaluation
model = GPT2LMHeadModel.from_pretrained("distilgpt2")
model.eval()

caleb-kaiser

# predictor.py

def predict(sample, metadata):
    indexed_tokens = tokenizer.encode(sample["text"])
    output = sample_sequence(model, metadata['num_words'], indexed_tokens, device=metadata['device'])
    return tokenizer.decode(
        output[0, 0:].tolist(), clean_up_tokenization_spaces=True, skip_special_tokens=True
    )

caleb-kaiser

# cortex.yaml

- kind: deployment
  name: text

- kind: api
  name: generator
  predictor:
    path: predictor.py
    metadata:

caleb-kaiser

from summarizer import Summarizer


class PythonPredictor:
    def __init__(self, config):
        self.model = Summarizer()

    def predict(self, payload):
        return self.model(payload["text"])

caleb-kaiser

transformers
spacy==2.1.3
bert-extractive-summarizer

caleb-kaiser

- kind: deployment
  name: text

- kind: api
  name: summarizer
  predictor:
    type: python
    path: predictor.py
  compute:
    mem: 4G

caleb-kaiser

import pickle
from dvc import api

ctx = {}

def init(model_path, metadata):
    ctx["model"] = pickle.loads(api.read(metadata["model_path"], metadata["dvc_repo"], mode="rb"))
    ctx["pipeline"] = pickle.loads(
        api.read(metadata["pipeline_path"], metadata["dvc_repo"], mode="rb")
    )