How to get model architecture/parameter names from the previous version
Hi. I have previously finetuned a Phi model, before the updates made https://huggingface.co/microsoft/phi-1_5/commit/d3ba318b780bfb92942c28853066fe4036d1b496. Now when I try to load my model using the previous code model = AutoModelForCausalLM.from_pretrained(model_path, use_flash_attention_2=False, trust_remote_code=True,device_map=device), I encounter an error that some weights are now used, like mentioned in https://huggingface.co/microsoft/phi-1_5/discussions/70. But I think mine issue is different since I try to load a previously trained model. Is there an easy way to fix this? Thanks!
Some attempt that I made: I have tried to load config and the model with a revision,config = AutoConfig.from_pretrained('microsoft/phi-1_5', revision='24f9ea14df973a49a0d87c16d04df88d90067468', trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained(model_path, config=config, use_flash_attention_2=False, torch_dtype=torch.float32, revision='24f9ea14df973a49a0d87c16d04df88d90067468', trust_remote_code=True, device_map=device) , however, this gives AttributeError: 'PhiConfig' object has no attribute 'attention_dropout'. It suggests that the config file is successfully retrieving the given revision, but the model initialization is not
i am facing The repository for microsoft/phi-1_5 contains custom code which must be executed to correctly load the model. You can inspect the repository content at https://hf.co/microsoft/phi-1_5. Please pass the argument trust_remote_code=True to allow custom code to be run. this error in my fine tuned model anyone suggest solution.