EleutherAI Builds open models, datasets, and evaluation tools so researchers can independently study large-scale AI, with work on interpretability and safety.
Products & media
Open-source framework that evaluates language models across academic benchmarks, supporting reproducible comparisons with shared tasks, prompts, and metrics.
Six-billion-parameter autoregressive language model with publicly released weights, supporting text generation experiments and downstream research on language models.
Diverse English-language training corpus combining multiple text sources, with replication code and documentation supporting language modeling research and dataset analysis.
Distributed library for training large autoregressive language models on GPUs, with public model weights including the GPT-NeoX-20B research release.
Open-source GPT-style language models and training code using Mesh TensorFlow, providing accessible checkpoints for language modeling research and experimentation.
Research model suite with intermediate training checkpoints and controlled data ordering, enabling studies of interpretability, scaling, and learning dynamics.
