Missing `transformers` requirement

[LysandreJik]

Dependencies don't mention transformers

diffusers/setup.py

Lines 78 to 91 in 850d434

	_deps = [
	"Pillow",
	"black~=22.0,>=22.3",
	"filelock",
	"flake8>=3.8.3",
	"huggingface-hub",
	"isort>=5.5.4",
	"numpy",
	"pytest",
	"regex!=2019.12.17",
	"requests",
	"torch>=1.4",
	"torchvision",
	]

However, importing from the quickstart fails without transformers installed:

In [1]: import torch

In [2]: from diffusers import UNetModel, DDPMScheduler
Einops is not installed
Einops is not installed
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
Input In [2], in <cell line: 1>()
----> 1 from diffusers import UNetModel, DDPMScheduler

File ~/Workspaces/Python/diffusers/src/diffusers/__init__.py:13, in <module>
 11 from .models.unet_grad_tts import UNetGradTTSModel
 12 from .pipeline_utils import DiffusionPipeline
---> 13 from .pipelines import DDIM, DDPM, GLIDE, LatentDiffusion, PNDM, BDDM
 14 from .schedulers import DDIMScheduler, DDPMScheduler, SchedulerMixin, PNDMScheduler
 15 from .schedulers.classifier_free_guidance import ClassifierFreeGuidanceScheduler

File ~/Workspaces/Python/diffusers/src/diffusers/pipelines/__init__.py:4, in <module>
 2 from .pipeline_ddpm import DDPM
 3 from .pipeline_pndm import PNDM
----> 4 from .pipeline_glide import GLIDE
 5 from .pipeline_latent_diffusion import LatentDiffusion
 6 from .pipeline_bddm import BDDM

File ~/Workspaces/Python/diffusers/src/diffusers/pipelines/pipeline_glide.py:27, in <module>
 24 from torch import nn
 26 import tqdm
---> 27 from transformers import CLIPConfig, CLIPModel, CLIPTextConfig, CLIPVisionConfig, GPT2Tokenizer
 28 from transformers.activations import ACT2FN
 29 from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling

ModuleNotFoundError: No module named 'transformers'

Is the end goal to have transformers as a hard or soft dependency?

[patrickvonplaten]

Thanks we should def make this a soft dependency

[anton-l]

Using dummy imports now instead of hard dependencies and try-catch blocks, thanks to Patrick ff885b0