Topic: code generation

Stefano Ermon discusses adapting diffusion models to language and code generation. Topics include discrete token spaces, parallel token generation, Mercury 2, and trade-offs with autoregressive models. The conversation examines training, serving and post-training challenges, controllable generation and possible directions for multimodal diffusion systems.

NEST [1] is a well-established open source simulator providing researchers in computational neuroscience with the ability to perform high-performance simulations of spiking neuronal networks. However, as the simulation kernel is written in C++ for performance reasons, this makes it challenging for researchers without a programming background to customize and extend the built-in neuron and synapse models. In order to satisfy both the need for high-performance simulation codes and a good user experience in terms of easy-to-use modeling of neurons and synapses, NESTML [2] was created as a domain-