The vocab size of the tokenizer 39410 is smaller than the vocab size of the base model 39424 This is not the intended use. Please check your model and tokenizer.
wtmshanbao opened this issue · 0 comments
wtmshanbao commented
我在合并ChatLaw13b和ziya时报错vocab_szie对应不上。我合并ziya模型时直接使用huggingface上的elinas/llama-13b-hf-transformers-4.29作为llama的基础模型,合并了modelscope上的Fengshenbang/Ziya-LLaMA-13B-v1。请问是哪个模型的版本与发布的模型不符合吗?