Docker image for the Text Generation Web UI: A Gradio web UI for Large Language Models with support for multiple inference backends.

Does Quantization Kill Interpretability? Scaling study across 5 models (124M-2.8B): RTN destroys induction heads in small models, GPTQ preserves them at all scales.

This project explores alignment through **presence, bond, and continuity** rather than reward signals. No RLHF. No preference modeling. Just relational coherence.

A Sign Forgery Detection System built with Python, utilizing image processing techniques to compare and verify handwritten signatures. This project demonstrates file handling, OpenCV-based analysis, and forgery detection logic in an intuitive interface.

TikiAI Natural Visual QA : UC Berkeley MIDS w251 Final Project

[ICLR'25 Spotlight] Min-K%++: Improved baseline for detecting pre-training data of LLMs

PyTorch implementation of normalization-free LLMs investigating entropic behavior to find desirable activation functions

Pythia is a versatile query format designed to facilitate the discovery of malicious infrastructure by seamlessly converting into the syntax of popular search engines.

an exercise in finetuning pythia suites of open source models for summarizing reviews of reserach papers

A re-implementation of the "Red Teaming Language Models with Language Models" paper by Perez et al., 2022

一套代码指令微调大模型

A simple MySQL connector for ArmA 3 build with Pythia

Python library for the Pythia platform.

Used as part of the CREST project by Nat Karmios, Nima Sahba, Carlos Barredo, and Lucas George.