TheBloke

/

dromedary-65B-lora-GGML

like 21

Sorry, didn't get a chance to do the model card yesterday. It'll be done shortly.

I need to re-do the GGMLs though as llama.cpp just push their breaking quant change

love you. you are doing the god's work

IBM Watson is back (alias Dromedary) and it beats GPT-4 at TruthfulQA-MC. It’s a variant of Constitutional AI, with LLaMa-65B as a base model and no RLHF or distillation from RLHF’d models. This seems good to me and may avoid the stubborn perverse-instantiation problems with RLHF.

Source: https://twitter.com/davidad/status/1656338102598930432

Source: https://arxiv.org/pdf/2305.03047.pdf#page=11