How to access model checkpoints without safetensor. I want to load the model in PyTorch Lightning about generative-models HOT 2 CLOSED

stability-ai commented on June 16, 2024

How to access model checkpoints without safetensor. I want to load the model in PyTorch Lightning

from generative-models.

Comments (2)

FurkanGozukara commented on June 16, 2024

diffuser files are already there

https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main

you can watch my this tutorial to see how my custom gradio loads

Stable Diffusion XL (SDXL) Locally On Your PC - 8GB VRAM - Easy Tutorial With Automatic Installer

from generative-models.

jenuk commented on June 16, 2024

There will be no distribution of non-safetensors files. However, that is not a problem, safetensors contain all important information and you can load safetensors easily into a pytorch lightning module, see for example our code here:

generative-models/sgm/models/diffusion.py

Lines 85 to 103 in 45c443b

 def init_from_ckpt( 

 self, 

 path: str, 

 ) -> None: 

 if path.endswith("ckpt"): 

 sd = torch.load(path, map_location="cpu")["state_dict"] 

 elif path.endswith("safetensors"): 

 sd = load_safetensors(path) 

 else: 

 raise NotImplementedError 

 missing, unexpected = self.load_state_dict(sd, strict=False) 

 print( 

 f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys" 

 ) 

 if len(missing) > 0: 

 print(f"Missing Keys: {missing}") 

 if len(unexpected) > 0: 

 print(f"Unexpected Keys: {unexpected}")

If you really need the older pytorch-based format, I would recommend that you convert the weights yourself, since it is not safe to open these files from other persons.

from generative-models.

How to access model checkpoints without safetensor. I want to load the model in PyTorch Lightning about generative-models HOT 2 CLOSED

Comments (2)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	def init_from_ckpt(
	self,
	path: str,
	) -> None:
	if path.endswith("ckpt"):
	sd = torch.load(path, map_location="cpu")["state_dict"]
	elif path.endswith("safetensors"):
	sd = load_safetensors(path)
	else:
	raise NotImplementedError

	missing, unexpected = self.load_state_dict(sd, strict=False)
	print(
	f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys"
	)
	if len(missing) > 0:
	print(f"Missing Keys: {missing}")
	if len(unexpected) > 0:
	print(f"Unexpected Keys: {unexpected}")