Skip to content

(Unofficial) building Hugging Face SmolLM-blazingly fast and remarkably powerful small language model with PyTorch implementation of grouped query attention (GQA)

Notifications You must be signed in to change notification settings

MyDarapy/SmolLM-experiments-with-grouped-query-attention

Repository files navigation

Improving Sub-billion Scale LLM Design Experiments

Some of the techniques used in the LLM pretraining design include:

  • Embedding Sharing
  • Grouped Query Design
  • SwiGLU Activations for the Multi Perceptron Layer
  • Intermidate blockwise weight sharing
chrome_7KTClIuUnl

About

(Unofficial) building Hugging Face SmolLM-blazingly fast and remarkably powerful small language model with PyTorch implementation of grouped query attention (GQA)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages