[Usage] Where to set the tokenizer in the sparse example?

[CharlesRiggins]

quantization_24_sparse_w4a16

I noticed that in the llama7b_sparse_w4a16 example, there doesn’t seem to be an option to specify a tokenizer. I’m curious if the example is using a LLaMA tokenizer by default. If so, this might be incorrect when other models are used.

[markurtz]

Good question; you can pass in the tokenizer with the tokenizer arg key. Otherwise it will default to loading the tokenizer used by the model.