Move ln_l to starter node and replace intermediate & finisher with "secondary" node

davmacario / MDI-LLM

Implementation of Model-Distributed Inference for Large Language Models, built on top of LitGPT

MIT License

3 stars 2 forks source link

Closed davmacario closed 7 months ago

davmacario commented 7 months ago

Now, just 2 node types are used - more scalable. Performance is the same!

This solves #15.