4 ms·
llama.cpp is just a frontend for the GGML tensor library, which required models be converted (optionally quantized) to GGML format. Of course people are workin
by summarity 4y ago
llama.cpp is just a frontend for the GGML tensor library, which required models be converted (optionally quantized) to GGML format.
Of course people are working on that for Dolly as well: https://huggingface.co/snphs/dolly-v2-12b-q4 https://huggingface.co/snphs/dolly-v2-12b-q4
- rodrigobellusci 4y agoBesides converting Dolly's weights to ggml it's also necessary to implement the model using ggml operators, right? Or does the ggml format also carry with it the architecture of the model?