notes.txt

model = AutoModelForCausalLM.from_pretrained("zhangfaen/Florence-2-large-ft", trust_remote_code=True).to(device)
processor = AutoProcessor.from_pretrained("zhangfaen/Florence-2-large-ft", trust_remote_code=True)

(Pdb++) sum([p.numel() for p in model.parameters()])
822693888
(Pdb++) sum([p.numel() for p in model.parameters() if p.requires_grad ])
822693888
(Pdb++) 


in /home/zhangfaen/miniconda3/envs/florence2-finetuning/lib/python3.11/site-packages/transformers/modeling_utils.py:3270
elif       os.path.isfile(                                                                                                                
3270                         os.path.join(pretrained_model_name_or_path, subfolder, _add_variant(WEIGHTS_NAME, variant))                                     
3271                     ):                                                                                                                                  
3272                         # Load from a PyTorch checkpoint                                                                                                
3273  ->                     archive_file = os.path.join(                                                                                                    
3274                             pretrained_model_name_or_path, subfolder, _add_variant(WEIGHTS_NAME, variant)                                               
3275                         )     
# WEIGHTS_NAME is pytorch_model.bin
(Pdb++) archive_file
'/home/zhangfaen/dev/Florence-2-large-ft/pytorch_model.bin'