Load all the weights for inference?
#8
by
leighgable
- opened
Hello!
Thanks for this super interesting fine-tuned model. I want to experiment with it. When I load it in a pipeline, or with AutoModelForCausalLM, I get the Some of the weights of the model checkpoint were not used when initializing PhiForCausalLM... namely, lm_head.linear.lora_A.default.weight as well as the b lora layer.
Do I need to load this model checkpoint in another manner? Or do some sort of merge_and_unload to merge the lora adapter into the model checkpoint? Any tips would be appreciated.
All the best and thanks again.