REVIEW 4 cited by
Training Plug-n-Play Knowledge Modules with Deep Context Distillation
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Dynamically integrating new or rapidly evolving information after (Large) Language Model pre-training remains challenging, particularly in low-data scenarios or when dealing with private and specialized documents. In-context learning and retrieval-augmented generation (RAG) face limitations, including their high inference costs and their inability to capture global document information. In this paper, we propose a way of modularizing knowledge by training document-level Knowledge Modules (KMs). KMs are lightweight components implemented as parameter-efficient LoRA modules, which are trained to store information about new documents and can be easily plugged into models on demand. We show that next-token prediction performs poorly as the training objective for KMs. We instead propose Deep Context Distillation: we learn KMs parameters such as to simulate hidden states and logits of a teacher that takes the document in context. Our method outperforms standard next-token prediction and pre-instruction training techniques, across two datasets. Finally, we highlight synergies between KMs and RAG.
Forward citations
Cited by 4 Pith papers
-
Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
Sleep-time Knowledge Seeding plus Dreaming lets LLMs expand capacity, distill fragile in-context memories into stable parameters, and self-improve without human labels.
-
Adapt Once, Thrive with Updates: Transferable Parameter-Efficient Fine-Tuning on Evolving Base Models
Trans-PEFT uses random FFN masking and layer dropping during fine-tuning so PEFT modules trained on an old base model transfer to a continually updated base model without retuning.
-
Cartridges: Lightweight and general-purpose long context representations via self-study
A per-corpus trained KV cache, called a Cartridge, matches full-context in-context learning quality on long-document benchmarks while using up to 38.6x less serving memory.
-
LoRA-Augmented Generation (LAG) for Knowledge-Intensive Language Tasks
LAG is a two-stage router that filters a 1,000-adapter LoRA library with Arrow and reranks with SpectR, outperforming the Arrow baseline and reaching 92.1% of its Oracle's performance on KILT tasks.
Discussion (0). Continue with ORCID to comment.