56 1 week ago

A custom Deepseek-coder:6.7b model with Deepseek-coder:1.3b as a speculative fill model to speed up inference. Primarily built for TabbyML usage.