From cb1f80704c2ebd3da332c121061f35815c5ccbc8 Mon Sep 17 00:00:00 2001 From: Daniel Bevenius Date: Tue, 16 Dec 2025 09:30:52 +0100 Subject: [PATCH] model-conversion : add note about verifying previous models This commit adds a note to the README in the model-conversion examples, advising developers to verify that previous versions of models pass logits verification before adding new models from the same family. --- examples/model-conversion/README.md | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/examples/model-conversion/README.md b/examples/model-conversion/README.md index 05d95d588b..8163b306b4 100644 --- a/examples/model-conversion/README.md +++ b/examples/model-conversion/README.md @@ -10,6 +10,13 @@ and in some cases perplexity checked of the quantized model. And finally the model/models need to the ggml-org on Hugging Face. This tool/example tries to help with this process. +> 📝 **Note:** When adding a new model from an existing family, verify the +> previous version passes logits verification first. Existing models can have +> subtle numerical differences that don't affect generation quality but cause +> logits mismatches. Identifying these upfront whether they exist in llama.cpp, +> the conversion script, or in an upstream implementation, can save significant +> debugging time. + ### Overview The idea is that the makefile targets and scripts here can be used in the development/conversion process assisting with things like: