philschmid

{
  "$schema": "https://schema.management.azure.com/schemas/2019-04-01/deploymentTemplate.json#",
  "contentVersion": "1.0.0.0",
  "metadata": {
    "_generator": {
      "name": "bicep",
      "version": "0.12.40.16777",
      "templateHash": "4423847801202994493"
    }
  },

philschmid

philschmid

philschmid

import tensorflow as tf 
from tensorflow.keras.optimizers import Adam
from transformers import TFAutoModelForSequenceClassification,AutoTokenizer
from datasets import load_dataset

# load model and tokenizer
model_id = "distilbert-base-uncased"
model = TFAutoModelForSequenceClassification.from_pretrained(model_id, num_labels=5)
tokenizer = AutoTokenizer.from_pretrained(model_id)

philschmid

# Setup Ubuntu
sudo apt update --yes
sudo apt upgrade --yes

# Get Miniconda and make it the main Python interpreter
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -b -p ~/miniconda 
rm ~/miniconda.sh
export PATH=~/miniconda/bin:$PATH

philschmid

# usage:
# deepspeed --num_gpus 8 bloom-ds-inference.py --name bigscience/bloom
#
# to run benchmarks:
# deepspeed --num_gpus 8 bloom-ds-inference.py --name bigscience/bloom --benchmark
#


# This is going to improve, but at the moment, the process is a bit cumbersome - we first use
# 1. use Deepspeed-ZeRO to instantiate the model on GPUs, w/o loading the checkpoints,

philschmid

philschmid

from sagemaker.huggingface.model import HuggingFaceModel
from sagemaker.serverless import ServerlessInferenceConfig
from sagemaker.serializers import DataSerializer
import sagemaker
import boto3

try:
    role = sagemaker.get_execution_role()
except ValueError:
    iam = boto3.client('iam')

philschmid

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: xgboost-trainer-
  annotations: {pipelines.kubeflow.org/kfp_sdk_version: 1.8.12, pipelines.kubeflow.org/pipeline_compilation_time: '2022-04-19T13:58:21.551241',
    pipelines.kubeflow.org/pipeline_spec: '{"description": "A trainer that does end-to-end
      distributed training for XGBoost models.", "inputs": [{"default": "gs://{{kfp-default-bucket}}",
      "name": "output", "optional": true}, {"default": "{{kfp-project-id}}", "name":
      "project", "optional": true}, {"default": "HALT_ON_ERROR", "name": "diagnostic_mode",
      "optional": true}, {"default": "5", "name": "rounds", "optional": true}], "name":

philschmid

from sagemaker.huggingface import HuggingFaceModel
from sagemaker.serializers import DataSerializer
import sagemaker

role = sagemaker.get_execution_role()
# Hub Model configuration. https://huggingface.co/models

hub = {
	'HF_MODEL_ID':'facebook/wav2vec2-base-960h',
	'HF_TASK':'automatic-speech-recognition'