Léo Innocenzi
Results
2
comments of
Léo Innocenzi
To anyone having this problem, make sure the folder or file name you're importing llama_cpp to is not named `llama_cpp` or `llama_cpp.py`.
Hello, After reviewing the [LoRA paper](http://arxiv.org/abs/2106.09685), it seems that the authors do not actually recommend using the q/k/v/o setting. Using the four weight types is doubling the size of the...