Skip to content

Support H-Net #33

@AndreasHolm

Description

@AndreasHolm

Goal

Support for H-Net (https://github.com/goombalab/hnet)

Design decisions

TBD

Future Extensions

  • Solve batching
  • (Distilled models from Token-based to Byte-based)

Benefits

  • Tokenizer-free modelling
  • Adds a foundation for future byte-level models

Definition of Done

  • Model train on test dataset
  • Generation load model from checkpoint and generate from prompt
  • Example small runnable generation example

Why Merge This?

  • Immediate value unlocks byte-level training and generation

Metadata

Metadata

Assignees

No one assigned

    Labels

    experimentDescription of an experiment to be performed

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions