Arcee AI's April 2024 tutorial tested whether its MergeKit toolkit could combine the base and instruction-tuned versions of Llama 3's eight-billion-parameter model into a single model. Using the SLERP merge method, the authors report that the merge succeeded and show uploading the result to the Hugging Face model hub. The post states its aim as showing MergeKit's compatibility with Llama 3 variants, so what it establishes is that one merge ran on this pair of models. The tutorial reports no evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → showing that the merged model outperformed either input, and it does not test other merge methods or model sizes.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
