Memristor compute-in-memory-based retrieval augemented generation
Year: 2025
Type: article
Abstract: The code of the paper "Fully integrated memristor compute-in-memory for massively parallel retrieval augmentation of large language models". Simulator: Evaluation of recall performance for digital-based and memristor-based retrieval under scaling. PubMed serves as the document corpus, while BioASQ and PubMedQA are used as query sets. The code includes quantization and mapping procedures to mapping both corpus and query vectors onto the memristor chip. RAG_BioASQ: Code implementation of the retrieval-augmented generation (RAG) task using the BioASQ test set. RAG_PubmedQA: Code implementation of the retrieval-augmented generation (RAG) task using the PubmedQA test set. Benchmark_CPU: Code for benchmarking CPU-based retrieval in terms of latency, throughput, and energy efficiency.
Author: Li, Xueqi
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7077275293
FWCI:
Cites:
Cited by:
Related to: