9 ms·

Maybe. Goliath 120B took two different llama variants and interwove the layers. Surprisingly Goliath 120B quantized to 2bit is outperforming llama 70B 4bit in m

• ghotli 3y ago
  Do you happen to have a link to where that interwoven layers bit is described? As far as I can tell it's not clear on the model cards.
  • valine 3y ago
    The model page is the only info I’ve found on it. As far as I can tell there’s no paper published on the technique. In the “Merge Process” section they at least give the layer ranges. https://huggingface.co/alpindale/goliath-120b https://huggingface.co/alpindale/goliath-120b
    • ghotli 3y ago
      Ah, actually reviewing that more closely I found a link to it in the acknowledgements. https://github.com/cg123/mergekit https://github.com/cg123/mergekit