Overview
Learn how to run inference on the massive BLOOM 176B language model using AWS's most powerful and expensive ML instance in this advanced technical tutorial. Explore implementation strategies for large language models, including model parallelism options like HuggingFace's accelerate pipeline and DeepSpeed tensor parallelism. Discover how to leverage Deep Java Library serving with PyTorch, use Language Model Interface Docker containers from Amazon ECR, and optimize for extremely low inference latency. Follow along with a Jupyter notebook demonstration showing BLOOM 176B model inference on AWS ml.p4de.24xlarge instances, comparing it with Flan-T5-XXL, and examining three different approaches for running LLMs on GPUs. Intended for experienced practitioners comfortable with cloud infrastructure costs, this deep technical dive provides hands-on guidance for implementing inference with trillion-parameter scale models using cutting-edge AWS technology and DeepSpeed optimizations.
Syllabus
BLOOM 176B vs Flan-T5-XXL
More Power!
3 Options to run LLMs on GPU
ipynb SageMaker DeepSpeed Container
Taught by
Discover AI