Open kousun12 opened 1 month ago
Hi @kousun12 , thanks for your interest in VILA! For the first question, I am wondering what is the version of your transformers, and how do you install VILA? The model arch llava_llama
should be already defined if you have installed VILA and the right version of transformers. For the second question, tokenizer.json
is here, under the llm
folder rather than the root of the model. You may need to modify the code for loading VILA to solve the problem. And please note that you need to make sure that the VILA model is served with newest TinyChat backends.
I'm wondering if I can get an easier pipeline by loading the awq weights with vllm:
The first issue seems to be that the
config.json
is trying to use a model type calledllava_llama
, whichtransformers
doesn't know about.if I change the type in
config.json
to justllava
I get:Which seems to suggest that the LLama tokenizer isn't in the llm directory? Do we need a
tokenizer.json
in the repo? Even if I add that, it seems to have trouble loading the tokenizer.