load

app.py

@@ -72,6 +72,10 @@ class FluxEditor:
         self.ae.eval()
         self.model.eval()
 
+        print(f"Inital_t5_device: {self.t5.hf_module.device}")
+        print(f"Inital_clip_device: {self.clip.hf_module.device}")
+        print(f"Inital_flow_model: {self.model.img_in.weight.device}")
+
         # clear history
         if os.path.exists("history_gradio/history.safetensors"):
             os.remove("history_gradio/history.safetensors")

src/flux/sampling.py

@@ -38,6 +38,9 @@ def prepare(t5: HFEmbedder, clip: HFEmbedder, img: Tensor, prompt: str | list[st
     if vec.shape[0] == 1 and bs > 1:
         vec = repeat(vec, "1 ... -> bs ...", bs=bs)
 
+    print(f"prepare t5 embedding: {txt}")
+    print(f"prepare clip embedding: {vec}")
+
     return {
         "img": img,
         "img_ids": img_ids.to(img.device),

src/flux/util.py

@@ -132,7 +132,7 @@ def load_flow_model(name: str, device: str | torch.device = "cuda", hf_download:
 def load_t5(device: str | torch.device = "cuda", max_length: int = 77) -> HFEmbedder:
     # max length 64, 128, 256 and 512 should work (if your sequence is short enough)
     return HFEmbedder("director432/Flux1-T5Encoder", max_length=max_length, is_clip=False, torch_dtype=torch.bfloat16).to(device)
-    #return HFEmbedder("/homedata/HuggingFace/black-forest-labs/FLUX.1-dev/text_encoder_2", max_length=max_length, is_clip=False, torch_dtype=torch.bfloat16).to(device)
+    #return HFEmbedder("/homedata/HuggingFace/black-forest-labs/print.1-dev/text_encoder_2", max_length=max_length, is_clip=False, torch_dtype=torch.bfloat16).to(device)
 
 
 def load_clip(device: str | torch.device = "cuda") -> HFEmbedder: