CacheRoute is an innovative LLM scheduling scheme dedicated to enabling flexible KV cache reuse across LLM systems, improving task performance and system efficiency.
ENGRAFT: write facts into an LLM's n-gram (Engram) memory table as a removable overlay: 100 facts in a 9 MB file, no weights touched, verified on llama.cpp GGUF (Qwen3.8-Flash-Next).
[NLPCC 2022] Kformer: Knowledge Injection in Transformer Feed-Forward Layers
A bilingual awesome list for VLM/MLLM knowledge injection research: benchmarks, papers, tools, resources, and ecosystem updates.
📚 Code for my master's thesis "Investigating Knowledge Injection Approaches for Research Field Classification of Scholarly Articles".
[EACL 2027] Are We Done with UMLS Injection into Language Models for Medical Question Answering? Systematic review of 25 studies and controlled experiments on how UMLS reaches language models.
Modular knowledge injection into frozen LLM backbones. Measured on Qwen3-4B / Qwen2.5-3B / SmolLM2-1.7B: a 64 to 128 sample phase transition, and a 2,560-parameter module form with no measurable drift on a held task.
We optimize a compact latent state (frozen weights) to force failed multi-hop chains to output the missing answer D. 5 pre-registered controls show it simply injects D: carries it without the code-fact, leaves intermediates invisible, inert to hop corruption, and doesn’t transfer. No latent composition at 3B (Llama-3.2-3B, Qwen2.5-3B).
A from-scratch PyTorch course: inject new knowledge into a teacher LLM, then distill it into a much smaller student — measured by cloze recall, laptop-sized.
Command-line interface for interacting with a modular RAG system: query, ingest, and manage knowledge pipelines.
MCP server that injects structured professional domain expertise into AI agents — accounting, legal, medical, Islamic finance, real estate, and more
When does grounded sensorimotor knowledge help a text model? A controlled cross-construct study of Lancaster sensorimotor norms on emotion classification and MBTI prediction. MSc dissertation.
To associate your repository with the knowledge-injection topic, visit your repo's landing page and select "manage topics."