Essay

Evaluating an LLM Serving Strategy for Different Use Cases

An engineering team is deploying a large language model for two distinct applications: (1) a real-time conversational chatbot where users expect consistently fast replies to short questions, and (2) an offline document analysis service where maximizing the number of long documents processed per day is the primary goal. The team is considering an inference serving strategy that maximizes hardware utilization by always prioritizing the initial, computationally heavy processing of newly arrived long documents over generating the next token for shorter requests already in progress. Evaluate the suitability of this strategy for each of the two applications. Justify your reasoning based on the performance trade-offs involved.

0

1

Updated 2025-10-02

Contributors are:

Who are from:

Tags

Ch.5 Inference - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Evaluation in Bloom's Taxonomy

Cognitive Psychology

Psychology

Social Science

Empirical Science

Science