Skip to main content

> ML_LIBRARY // DJL_v1.0

Deep Java Library (DJL)

Amazon Web Services (AWS) — AWS's engine-agnostic deep learning framework for Java, Scala, and Kotlin.

non-python-ecosystemsv0.29.0Apache-2.0qualified
Official DocumentationSource Repository

Model Training

Supported
Accelerators:
CPUCUDA
Distributed Training:Yes

Model Inference

Supported
Inference Accelerators:
CPUCUDA
Deployment Targets:server

What It Does

  • +Engine-agnostic Java API bridging to native PyTorch, TensorFlow, ONNX Runtime, TensorRT, and fastText C++ engines
  • +DJL Serving: high-performance model serving container optimized for multi-GPU LLMs and PyTorch models on Kubernetes
  • +Seamless integration running deep learning inference inside Apache Spark and Apache Flink jobs
  • +Hugging Face model hub integration pulling and executing tokenizers directly in Java

What It Does Not Do

  • -Provide a pure Python data science interactive shell
  • -Run on edge microcontrollers without a full Java Virtual Machine runtime
  • -Replace relational SQL query planning

>Suitable Work Types

  • Running deep learning inference directly inside Apache Spark or Apache Flink data pipelines in Java/Scala
  • Serving PyTorch and Hugging Face transformer models inside enterprise Java microservices without Python processes
  • High-throughput enterprise model serving on AWS EKS with DJL Serving

>Unsuitable Work Types

  • Research workflows where fast Python prototyping is preferred
  • Simple linear regressions where standard Java libraries like Smile are simpler
Data Residency Implications

Runs locally inside your private enterprise JVM instances. Zero telemetry sent to AWS by default.

Security Considerations

Apache-2.0 license. Developed and backed by AWS open-source team.

Operational Profile & Known Limitations

Maturity:mature
Learning Curve:moderate
Ops Complexity:low
Cost Tier:free-oss
> Known Limitations:
  • Native C++ engine binaries (libtorch, onnxruntime) must be downloaded or packaged for the exact host OS and architecture (x86_64 vs aarch64).

Associated Incident Patterns (Incidentpedia)

Enforce safeguards and monitoring to guard against these documented real-world failure modes:

cuda-driver-torch-mismatch

> Primary Evidence & Benchmark Citations

Deep Java Library Documentationofficial-docs • >=0.28.0, <=0.29.x
2026-09-25HIGH