Train a LLM from Scratch(github.com)
3 pointsby linhns6 hours ago1 comment
  • subtick6 hours ago
    Curious — how did you handle training stability early on? Was convergence an issue without heavy tuning?