Any plans to create or release smaller models (3B, 1B)?

[sozforex]

Hi! Do you have any plans to create or release smaller Code Llama models?
They may be of use to speed-up inference via speculative sampling (ggerganov/llama.cpp#2926).

[jspisak]

thanks for the request - this is definitely something the community has been asking for. We are in roadmap planning now and we will take this into account. Btw love the GGML work!