Skip to content

CRoM v1.0.1 – Context Rot Mitigation Core Release

Compare
Choose a tag to compare
@Flamehaven Flamehaven released this 06 Sep 05:44
· 7 commits to main since this release

🧠 CRoM v1.0.1 – Context-Rot Mitigation for Efficient LLMs

Release Date: 2025-09-06
Tag: v1.0.1


🚀 What's New

First functional release of the CRoM system — a context efficiency layer for Retrieval-Augmented LLM systems. This version brings token-aware injection, cross-encoder fallback, and full FastAPI orchestration.


🔧 Features

  • budget_packer.py – Smart prompt packing under token budgets
  • Cross-Encoder Fallback – Embedding-based scoring when LLMs are uncertain
  • Capsule Logger – Transparent logging of injected/pruned context
  • FastAPI Interface – Live endpoint access for simulation, injection, and logging

⚠️ Known Issues

  • Persistent vector cache (FAISS) not yet implemented
  • Evaluation tests are internal only
  • Single-thread fallback slows batch performance

🔭 Roadmap

  • Vector memory cache
  • Context summarizer & history pruning
  • Performance benchmarking suite

🔌 Install & Run

git clone https://github.com/Flamehaven/CRoM-Context-Rot-Mitigation--EfficientLLM.git
cd CRoM-Context-Rot-Mitigation--EfficientLLM
pip install -r requirements.txt
uvicorn app.main:app --reload