Update app.py

app.py

@@ -23,10 +23,10 @@ import warnings
 import safetensors.torch
 
 
-#cache_path = path.join(path.dirname(path.abspath(__file__)), "models")
-#os.environ["TRANSFORMERS_CACHE"] = cache_path
-#os.environ["HF_HUB_CACHE"] = cache_path
-#os.environ["HF_HOME"] = cache_path
+cache_path = path.join(path.dirname(path.abspath(__file__)), "models")
+os.environ["TRANSFORMERS_CACHE"] = cache_path
+os.environ["HF_HUB_CACHE"] = cache_path
+os.environ["HF_HOME"] = cache_path
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
@@ -46,23 +46,25 @@ dtype = torch.bfloat16
 pipe = FluxWithCFGPipeline.from_pretrained("ostris/OpenFLUX.1", torch_dtype=dtype).to("cuda")
 pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to("cuda")
 #pipe.num_single_layers="0"
-pipe.transformer_chunk_size="0"
+#pipe.transformer_chunk_size="0"
 #model.pooled_projections="(_, 1)[0]"
-pipe.transformer_pooled_projections_dim="(batch_size, 0)"
+#pipe.transformer_pooled_projections_dim="(batch_size, 0)"
 
 pipe.to("cuda")
-#clipmodel = 'norm'
-#if clipmodel == "long":
-#    model_id = "zer0int/LongCLIP-GmP-ViT-L-14"
-#    config = CLIPConfig.from_pretrained(model_id)
-#if clipmodel == "norm":
-#    model_id = "zer0int/CLIP-GmP-ViT-L-14"
-#    config = CLIPConfig.from_pretrained(model_id)
-#clip_model = CLIPModel.from_pretrained(model_id, torch_dtype=torch.bfloat16, config=config, ignore_mismatched_sizes=False).to("cuda")
-#clip_processor = CLIPProcessor.from_pretrained(model_id, padding="max_length", ignore_mismatched_sizes=False, return_tensors="pt", truncation=True)
-#pipe.tokenizer = clip_processor.tokenizer
-#pipe.text_encoder = clip_model.text_model
-#pipe.text_encoder.dtype = torch.bfloat16
+clipmodel = 'norm'
+if clipmodel == "long":
+    model_id = "zer0int/LongCLIP-GmP-ViT-L-14"
+    config = CLIPConfig.from_pretrained(model_id)
+    maxtokens = 77
+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.text_encoder.dtype = torch.bfloat16
 torch.cuda.empty_cache()
 
 MAX_SEED = 2**32-1