A company uses an AWS Lambda function to perform natural language processing (NLP) tasks.

Requirement Summary:

NLP Lambda function with a large pre-trained model

Lambda layer became 8.7 GB → Exceeds AWS limits

Function returns RequestEntityTooLargeException

Need: High-performing, portable, low initialization time

Important AWS Limits:

Lambda Layers size limit (combined across all layers): 250 MB (unzipped)

Deployment package size (unzipped): 250 MB

Lambda container image support allows up to 10 GB image size

Evaluate Options:

A: Store model in S3 and load during execution

Leads to cold start latency every time

Model loading from S3 is slower and not suitable for real-time NLP

Not optimal for performance

B: Use EFS mounted to Lambda

⚠️ Valid for large models, but adds latency during cold start as model loads from EFS

Requires EFS setup, VPC, and has added network I/O overhead

Still slower than bundling in container image

C: Split into five Lambda layers

Still violates the total layer size limit of 250 MB (unzipped)

You cannot exceed that even with multiple layers

D: Use Docker container image

Allows bundling up to 10 GB of dependencies and models

High portability and performance

Avoids latency of downloading models at runtime

Ideal for scientific/NLP models

Lambda container image support: https://docs.aws.amazon.com/lambda/latest/dg/images-create.html

Lambda limits: https://docs.aws.amazon.com/lambda/latest/dg/gettingstarted-limits.html

Using large models with Lambda: https://aws.amazon.com/blogs/machine-learning/deploying-large-machine-learning-models-on-aws-lambda-with-container-images/