a

app.py

@@ -29,7 +29,10 @@ if HF_TOKEN: login(token=HF_TOKEN)
 cp_dir = os.getenv('CHECKPOINT_DIR', 'checkpoints')
 snapshot_download("Djrango/Qwen2vl-Flux", local_dir=cp_dir)
 hf_hub_download(repo_id="TheMistoAI/MistoLine", filename="MTEED.pth", subfolder="Anyline", local_dir=f"{cp_dir}/anyline")
-shutil.move("checkpoints/anyline/Anyline/MTEED.pth", f"{cp_dir}/anyline")
+try:
+    shutil.move("checkpoints/anyline/Anyline/MTEED.pth", f"{cp_dir}/anyline")
+except:
+    print("anyline fail")
 snapshot_download("depth-anything/Depth-Anything-V2-Large", local_dir=f"{cp_dir}/depth-anything-v2")
 snapshot_download("facebook/sam2-hiera-large", local_dir=f"{cp_dir}/segment-anything-2")
 # https://github.com/facebookresearch/sam2/issues/26

modelmod.py

@@ -122,10 +122,10 @@ class FluxModel:
         self.connector.to(self.dtype).to(self.device)
 
         # Text encoders initialization
-        self.tokenizer = CLIPTokenizer.from_pretrained(MODEL_PATHS['flux'], subfolder="tokenizer")
-        self.text_encoder = CLIPTextModel.from_pretrained(MODEL_PATHS['flux'], subfolder="text_encoder")
-        self.text_encoder_two = T5EncoderModel.from_pretrained(MODEL_PATHS['flux'], subfolder="text_encoder_2", **self.qkwargs)
-        self.tokenizer_two = T5TokenizerFast.from_pretrained(MODEL_PATHS['flux'], subfolder="tokenizer_2")
+        self.tokenizer = CLIPTokenizer.from_pretrained(MODEL_PATHS['flux'], subfolder="tokenizer", token=HF_TOKEN)
+        self.text_encoder = CLIPTextModel.from_pretrained(MODEL_PATHS['flux'], subfolder="text_encoder", token=HF_TOKEN)
+        self.text_encoder_two = T5EncoderModel.from_pretrained(MODEL_PATHS['flux'], subfolder="text_encoder_2", token=HF_TOKEN, **self.qkwargs)
+        self.tokenizer_two = T5TokenizerFast.from_pretrained(MODEL_PATHS['flux'], subfolder="tokenizer_2", token=HF_TOKEN)
 
         self.text_encoder.requires_grad_(False).to(self.dtype).to(self.device)
         #self.text_encoder_two.requires_grad_(False).to(self.dtype).to(self.device)
@@ -139,9 +139,9 @@ class FluxModel:
         self.t5_context_embedder.to(self.dtype).to(self.device)
 
         # Basic components
-        self.noise_scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(MODEL_PATHS['flux'], subfolder="scheduler", shift=1)
-        self.vae = AutoencoderKL.from_pretrained(MODEL_PATHS['flux'], subfolder="vae")
-        self.transformer = FluxTransformer2DModel.from_pretrained(MODEL_PATHS['flux'], subfolder="transformer", **self.qkwargs)
+        self.noise_scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(MODEL_PATHS['flux'], subfolder="scheduler", shift=1, token=HF_TOKEN)
+        self.vae = AutoencoderKL.from_pretrained(MODEL_PATHS['flux'], subfolder="vae", token=HF_TOKEN)
+        self.transformer = FluxTransformer2DModel.from_pretrained(MODEL_PATHS['flux'], subfolder="transformer", token=HF_TOKEN, **self.qkwargs)
 
         self.vae.requires_grad_(False).to(self.dtype).to(self.device)
         #self.transformer.requires_grad_(False).to(self.dtype).to(self.device)