Pin compiler versions (Dockerfile with specific gcc/clang versions and flags), commit a lockfile of LLM model versions used, version the dataset itself with a hash. Today running the same model a year from now will give different results because the model itself changed under the same ID.
Pin compiler versions (Dockerfile with specific gcc/clang versions and flags), commit a lockfile of LLM model versions used, version the dataset itself with a hash. Today running the same model a year from now will give different results because the model itself changed under the same ID.