Summary of Marlin: Mixed-precision Auto-regressive Parallel Inference on Large Language Models, by Elias Frantar et al.
MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models
by Elias Frantar, Roberto L. Castro, Jiale Chen, Torsten Hoefler, Dan Alistarh
First submitted to arxiv on: 21 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates the feasibility of applying weight quantization techniques on Large Language Models (LLMs) in batched settings, where multiple parallel clients perform inference concurrently. Weight quantization has been shown to reduce model size and achieve speedups for single-user inference, but its effectiveness in batched settings is unclear. The authors aim to design GPU kernels that remain memory-bound while supporting the increased compute requirements of batched workloads. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper explores whether weight quantization can also bring speedup benefits when processing multiple inputs simultaneously, which is crucial for real-world applications where many users perform inference concurrently. |
Keywords
» Artificial intelligence » Inference » Quantization
