Lambda for ML Inference

123456789101112
# Use AWS Lambda Python base image for ARM64
FROM public.ecr.aws/lambda/python:3.11-arm64

# Install PyTorch CPU-only (saves ~1GB vs CUDA version)
RUN pip install torch==2.1.0 --index-url https://download.pytorch.org/whl/cpu

# Install additional ML dependencies
RUN pip install transformers==4.35.0 \
    numpy==1.24.0 \
    scipy==1.11.0 \
    --no-cache-dir

123456789101112
import json
import os
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

# Global variables for model persistence across warm invocations
model = None
tokenizer = None
MODEL_PATH = os.environ.get('MODEL_PATH', '/var/task/model_weights')
CACHE_PATH = '/tmp/model_cache'

def load_model():

123456789101112
# Stage 1: Build dependencies
FROM public.ecr.aws/lambda/python:3.11 as builder

WORKDIR /var/task

# Install build dependencies
RUN yum install -y gcc gcc-c++ make

# Copy and install requirements
COPY requirements.txt .
RUN pip install --no-cache-dir \
    --target /var/task/dependencies \

123456789101112
import json
import os
import boto3
from functools import lru_cache

# Global scope - loaded once per execution environment
print("Cold start: Loading model...")

# Lazy loading pattern with caching
@lru_cache(maxsize=1)
def get_model():
    """Load model once and cache in memory."""

123456789101112
import json
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from aws_lambda_powertools import Logger, Tracer, Metrics
from aws_lambda_powertools.metrics import MetricUnit
from aws_lambda_powertools.utilities.typing import LambdaContext

logger = Logger()
tracer = Tracer()
metrics = Metrics()

# Load model at module level (runs during init)

123456789101112
# Stage 1: Download and prepare model
FROM python:3.11-slim as model-builder

WORKDIR /model
RUN pip install transformers torch --no-cache-dir

# Download model during build
RUN python -c "
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = 'distilbert-base-uncased-finetuned-sst-2-english'

123456789101112
import * as cdk from 'aws-cdk-lib';
import * as lambda from 'aws-cdk-lib/aws-lambda';
import * as apigateway from 'aws-cdk-lib/aws-apigateway';
import * as cloudwatch from 'aws-cdk-lib/aws-cloudwatch';
import * as applicationautoscaling from 'aws-cdk-lib/aws-applicationautoscaling';

export class MLInferenceStack extends cdk.Stack {
  constructor(scope: cdk.App, id: string, props?: cdk.StackProps) {
    super(scope, id, props);

    // ML Inference Lambda with container image
    const inferenceFunction = new lambda.DockerImageFunction(this, 'InferenceFunction', {