Closed NohTow closed 1 month ago
Since the tokenizer has not the same number of tokens than the base model embeddings, we need to allow resizing it.
I insist, but maybe it's better we get ride of anything that is set as default by the base model and explicitly define everything in the config.
Since the tokenizer has not the same number of tokens than the base model embeddings, we need to allow resizing it.
I insist, but maybe it's better we get ride of anything that is set as default by the base model and explicitly define everything in the config.