I used https://huggingface.co/sambanovasystems/SambaLingo-Hungarian-Chat-70B q40. Conversion was successful but it reported incompatible tokenizer.

Commented out the 'if (maxTokenLength < 1 || vocabSize != modelVocabSize)' expression and it worked as it should be. So maybe sometimes the tokenizer and the model have not the same size of dictionary.

Maybe a warning would be enough in these cases.