Virgil Crypto is a high-level cryptographic library that allows you to perform all necessary operations for secure storing and transferring data and everything required to become HIPAA and GDPR compliant. Crypto Library is written in C++, suitable for mobile and server platforms and supports bindings with: Swift, Obj-C, Java (Android), С#/.NET, …
[ICLR'25 Spotlight] Min-K%++: Improved baseline for detecting pre-training data of LLMs
This library is designed to be small, flexible and convenient wrapper for a variety crypto algorithms. So it can be used in a small micro controller as well as in a high load server application.
一套代码指令微调大模型
A re-implementation of the "Red Teaming Language Models with Language Models" paper by Perez et al., 2022
Pythia is a versatile query format designed to facilitate the discovery of malicious infrastructure by seamlessly converting into the syntax of popular search engines.
Machine Learning Experiment Monitoring Platform
A small LLM built from scratch in PyTorch. Loads GPT-2, Llama (SmolLM2), Qwen2.5 and Pythia weights with outputs identical to Hugging Face. LoRA fine-tuning, KV cache, Kaggle/Colab notebook and a Streamlit chatbot.
Virgil Pythia SDK allows developers to implement Pythia protocol to create breach-proof passwords, immune to offline and online attacks.
Visualise particle event tree from High-Energy Monte Carlo generators
Virgil Crypto library is a small, flexible and convenient wrapper for a variety of cryptographic algorithms. Includes asymmetric key generation, encryption/decryption of data, generation/verification of digital signatures, Pythia protocol and Double Ratchet protocol implementation.
Virgil Pythia Library
Pythia 8 Docker image with Python 3 and the HEP simulation stack
Virgil Pythia SDK allows developers to implement Pythia protocol to create breach-proof passwords, immune to offline and online attacks.
Full-stack web application built at CDAC Patna — connects students with campus resources | HTML, CSS, JavaScript, Python, MySQL | Team Leader
TMLR 2026 | Mechanistic interpretability: attention-head binding (EB*) as a marker of concept emergence. 7 models, 5 architectures (Pythia 160M–2.8B, OLMo-1B, CRFM GPT-2, SmolLM3-3B, Qwen2.5-1.5B), 41 terms.
Code for "Learning to Read Out: Unembedding Dynamics in Language Model Pretraining" (NeurIPS 2026, arXiv:2610.21025). Parameter-trajectory crosscoders on the unembedding across pretraining checkpoints.
Replacing LayerNorm with Dynamic Tanh (DyT) in DistilGPT2 + LoRA, evaluated on RE-WILD, Alpaca, and ShareGPT.
To associate your repository with the pythia topic, visit your repo's landing page and select "manage topics."