EleutherAI

lm-evaluation-harness

A framework for few-shot evaluation of language models.

gpt-neox

An implementation of model parallel autoregressive transformers on GPUs, based on the Megatron and DeepSpeed libraries

pythia

The hub for EleutherAI's work on interpretability and learning dynamics

cookbook

Deep learning for dummies. All the practical details and useful utilities that go into working with real models.

sparsify

Sparsify transformers with SAEs and transcoders

concept-erasure

Erasing concepts from neural representations with provable guarantees

elk

Keeping language models honest by directly eliciting knowledge encoded in their activations.

delphi

Delphi was the home of a temple to Phoebus Apollo, which famously had the inscription, 'Know Thyself.' This library lets language models know themselves through automated interpretability.

DeeperSpeed

DeepSpeed is a deep learning optimization library that makes distributed training easy, efficient, and effective.

nanoGPT-mup

The simplest, fastest repository for training/finetuning medium-sized GPTs.

aria

semantic-memorization

aria-amt

Efficient and robust implementation of seq-to-seq automatic piano transcription.

training-jacobian

mdl

Minimum Description Length probing for neural network representations

polyapprox

Closed-form polynomial approximations to neural networks

transformer-reasoning

Experiments in transformer knowledge and reasoning

cupbearer

A library for mechanistic anomaly detection

tyche

Precisely estimating the volume of basins in neural net parameter space corresponding to interpretable behaviors

ccs

clearnets

ngrams-across-time

website

New website for EleutherAI based on Hugo static site generator

sae_overlap

Acompanying code for our research on SAE feature overlap when trained on different seeds.

aria-utils

MIDI tokenizers and pre-processing utils.

rtopk

https://github.com/xiexi51/RTopK PyTorch wrapper

mixture-of-depths

open-r1

Fully open reproduction of DeepSeek-R1

POSER

Poser: Unmasking Alignment Faking LLMs by Manipulating Their Internals

TransformerEngine

A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit floating point (FP8) precision on Hopper and Ada GPUs, to provide better performance with lower memory utilization in both training and inference.