NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals
This democratizes access to foundation model internals for researchers, addressing a growing gap in AI literature, though it is incremental as it builds on existing inference and execution frameworks.
The authors tackled the problem of limited access to the internals of large neural networks like LLMs by introducing NNsight and NDIF, which provide transparent and efficient tools for scientific study, enabling research on huge models without individual hosting costs.
We introduce NNsight and NDIF, technologies that work in tandem to enable scientific study of the representations and computations learned by very large neural networks. NNsight is an open-source system that extends PyTorch to introduce deferred remote execution. The National Deep Inference Fabric (NDIF) is a scalable inference service that executes NNsight requests, allowing users to share GPU resources and pretrained models. These technologies are enabled by the Intervention Graph, an architecture developed to decouple experimental design from model runtime. Together, this framework provides transparent and efficient access to the internals of deep neural networks such as very large language models (LLMs) without imposing the cost or complexity of hosting customized models individually. We conduct a quantitative survey of the machine learning literature that reveals a growing gap in the study of the internals of large-scale AI. We demonstrate the design and use of our framework to address this gap by enabling a range of research methods on huge models. Finally, we conduct benchmarks to compare performance with previous approaches. Code, documentation, and tutorials are available at https://nnsight.net/.