AdaMem: Adaptive Memory Token Allocation for Soft Compression in Retrieval-Augmented Generation

Researchers proposed AdaMem, a relevance-guided soft-compression framework for retrieval-augmented generation that maps learned passage-relevance estimates to a query-dependent allocation of a fixed memory-token budget, according to an arXiv paper (arXiv:2609.22100v1). Across six open-domain QA benchmarks, AdaMem outperformed the uniform-allocation soft-compression baseline OSCAR, improving sub-string match by up to 3.2 points (5.5%) at 16x compression with an average relative gain of 3.4%, and by an average relative gain of 14.6% at 64x compression, including a maximum of 9.8 points (19.7%) on PopQA. AdaMem matched uncompressed answer quality at up to 4x lower inference latency than full-context inference while retaining an efficiency profile comparable to the uniform-compression baseline.

arXiv:2609.22100v1 Announce Type: new Abstract: Retrieval-augmented generation RAG improves language models with retrieved evidence, but processing many long passages is costly and can introduce distracting information. Soft compression addresses this challenge by encoding passages as compact sequences of continuous memory embeddings before generation. However, existing methods typically assign each retained passage an identical number of memory embeddings, irrespective of its query-specific relevance. To address this, we propose AdaMem, a relevance-guided soft-compression framework that maps learned passage-relevance estimates to a query-dependent allocation of a fixed memory-token budget. A shared query-conditioned compressor produces both continuous passage memories and relevance scores in a single pass; a deterministic allocation rule assigns more memory tokens to higher-scoring passages and can omit low-scoring ones. Across six open-domain QA benchmarks, AdaMem consistently outperforms OSCAR the closely matched soft-compression baseline that uses uniform allocation as well as other soft-compression methods at matched memory budgets. Under standard 16$\times$ compression, AdaMem improves sub-string match by up to 3.2 points 5.5% over uniform allocation baseline, with an average relative gain of 3.4%; under aggressive 64$\times$ compression the average relative gain grows to 14.6%, with a maximum of 9.8 points 19.7% on PopQA. AdaMem matches the answer quality of the uncompressed at up to 4$\times$ lower inference latency than full context baseline. AdaMem retains an efficiency profile comparable to the uniform-compression baseline, while achieving up to $4\times$ lower inference latency than full-context inference. Thus, relevance-guided memory allocation is particularly effective when retrieval pools are large and the available memory budget is tight.