slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
lat-defuse-rag-repair
A research repository that reproduces the Lat-Defuse sparse-autoencoder intervention method on Gemma-3-12B and mechanistically evaluates how it corrects RAG errors caused by misleading retrieved documents.
Stars
3
7-day growth
No data
Forks
0
Open issues
16
License
MIT
Last updated
2026-07-30
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It provides a faithful, code-independent adaptation of a peer-reviewed method, made auditable through a science contract, compatibility checks, and committed run reports.
Who it is for
- Mechanistic interpretability researchers
- Researchers studying RAG robustness and distractor effects
- Practitioners working with sparse autoencoders and activation steering
- Open-source contributors interested in reproducible ML research
Use cases
- Reproducing or adapting Lat-Defuse-style interventions on a modern 12B model
- Investigating which internal SAE features matter for distractor-induced RAG failures
- Developing and validating SAE-based activation repair methods
- Contributing to a research codebase with clear CI and experiment-tracking conventions
Strengths
- Reproduces a published ACL Findings method without copying its code
- Defines a science contract that pins down the numbers and rules for experiments
- Committed run reports make real-model experiment outputs auditable
- Compatibility failures raise exceptions by design, preventing silent correctness issues
Considerations
- Running the real model and SAE requires GPU access
- Data and results directories are not committed, so some artifacts must be generated locally
- The repository is a faithful adaptation, not the official upstream implementation
README quick start
lat-defuse-rag-repair
Mechanistic evaluation of SAE-based activation interventions for correcting distractor-induced errors in RAG.
Retrieval-augmented generation (RAG) feeds a language model retrieved documents next to a question. When one of those documents is misleading (a distractor), the model often abandons the correct answer it would otherwise give. Lat-Defuse is a published method that repairs this failure by adjusting the model's internal sparse-autoencoder features until the correct answer wins again. This repository builds a faithful adaptation of that method for Gemma-3-12B and then studies what the repair actually does inside the model.
The reference implementation is the official code of the ACL 2026 Findings paper "The Mechanics of Interference: Defusing Distractors in RAG via Sparse Autoencoder Interventions" by Christian Giannetti and coauthors: at commit eff2cdfe664bc38d9de9ec3fb047b25d471826df (MIT). We do not copy its code; we reproduce its behavior. See docs/upstream-notes.md.
Directory map
docs/ the science contract, upstream notes, dataset survey
configs/ experiment configuration files
scripts/ runnable entry points
data/ datasets and curated examples, never committed
results/ experiment outputs, never committed
reports/ tracked run reports, one per real run
src/latdefuse/ the package
data/ example schema, prompts, curated-example loader
model/ model loading, activation capture, activation replacement
sae/ SAE loading and compatibility checks
steering/ the Lat-Defuse optimizer and stopping rules
tests/ unit tests, runnable without a GPU
Quickstart
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
pytest -q
How we work
- Pick an open issue by label. Issues labeled "good first issue" need no GPU and no downloads.
- One issue, one focused pull request. CI (ruff, mypy, pytest) must be green before review.
- Issues labeled "needs-gpu" require a machine with the real model; everything else runs locally on a tiny test model.
- Read docs/science-contract.md before touching anything under
src/latdefuse/. It defines the numbers and rules the sci
Description
Mechanistic evaluation of SAE-based activation interventions for correcting distractor-induced errors in RAG.
Related repositories
Similar projects matched by category, topics, and programming language.
gavamedia
Featured
deltafin
Deltafin is a research project that runs the 2.8-trillion-parameter Mixture-of-Experts model Kimi K3 on a single Apple Silicon Mac (e.g., M1 Max with 64 GB) at about 16 seconds per token, using exact, reproducible inference with local or streaming expert loading.
jamesob
Featured
local-llm
A comprehensive guide for building and configuring a high-end local machine to run state-of-the-art LLMs, with detailed hardware choices, BIOS tuning, and Docker-based model serving.