Learn Before
Inference Bottleneck on Memory-Constrained Devices
Analyze this implementation and identify the primary resource constraint that will likely cause performance issues when a user provides a very long article. Explain why this specific approach leads to that particular constraint.
0
1
Tags
Ch.5 Inference - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Comparison of Processing in Chunked vs. Standard Prefilling
A large language model is tasked with processing a very long input document. To prepare for generating a response, it computes the Key-Value cache for the entire document in a single, large forward pass before any new tokens are produced. What is the most significant computational challenge or trade-off inherent to this 'all-at-once' approach?
A user submits a prompt to a large language model that uses a conventional inference process. Arrange the following stages in the correct chronological order, from receiving the prompt to generating the first new word.
Inference Bottleneck on Memory-Constrained Devices