Update app.py

app.py

@@ -8,9 +8,11 @@ from PIL import Image
 import spaces
 from diffusers import DiffusionPipeline, AutoPipelineForText2Image
 from diffusers import StableDiffusion3Pipeline, FlowMatchEulerDiscreteScheduler, SD3Transformer2DModel # pip install diffusers>=0.31.0
+from transformers import CLIPModel, CLIPProcessor, CLIPTextModel, CLIPTokenizer, CLIPConfig, T5EncoderModel, T5Tokenizer
 import copy
 import random
 import time
+from huggingface_hub import HfFileSystem, ModelCard
 from huggingface_hub import login, hf_hub_download
 import safetensors.torch
 from safetensors.torch import load_file
@@ -38,6 +40,24 @@ with open('loras.json', 'r') as f:
 #base_model = "stabilityai/stable-diffusion-3.5-large"
 pipe = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-3.5-large", torch_dtype=torch.bfloat16)
 
+clipmodel = 'norm'
+if clipmodel == "long":
+    model_id = "zer0int/LongCLIP-GmP-ViT-L-14"
+    config = CLIPConfig.from_pretrained(model_id)
+    maxtokens = 248
+if clipmodel == "norm":
+    model_id = "zer0int/CLIP-GmP-ViT-L-14"
+    config = CLIPConfig.from_pretrained(model_id)
+    maxtokens = 77
+clip_model = CLIPModel.from_pretrained(model_id, torch_dtype=torch.bfloat16, config=config, ignore_mismatched_sizes=True).to("cuda")
+clip_processor = CLIPProcessor.from_pretrained(model_id, padding="max_length", max_length=maxtokens, ignore_mismatched_sizes=True, return_tensors="pt", truncation=True)
+
+pipe.tokenizer = clip_processor.tokenizer
+pipe.text_encoder = clip_model.text_model
+pipe.tokenizer_max_length = maxtokens
+pipe.text_encoder.dtype = torch.bfloat16
+
+
 #pipe.transformer.to(memory_format=torch.channels_last)
 #pipe.vae.to(memory_format=torch.channels_last)