Update app.py

app.py

@@ -138,15 +138,31 @@ def load_models() -> str | None:
     try:
         import torch
         from diffusers import ComponentsManager
-
         from h3_split_blocks import MiniMaxH3GeneratorBlocks
-
+    
         lower_duration_floor()
+    
         manager = ComponentsManager()
         blocks = MiniMaxH3GeneratorBlocks()
-        print(f"[gen] loading {[c.name for c in blocks.expected_components]} from {MODEL_REPO} ...", flush=True)
-        pipe = blocks.init_pipeline(MODEL_REPO, components_manager=manager, collection="h3")
+    
+        print(
+            f"[gen] loading {[c.name for c in blocks.expected_components]} from {MODEL_REPO} ...",
+            flush=True,
+        )
+    
+        pipe = blocks.init_pipeline(
+            MODEL_REPO,
+            components_manager=manager,
+            collection="h3",
+        )
+    
         pipe.load_components(dtype=torch.bfloat16)
+        # Load TenStrip fine-tuned checkpoint into the transformer
+        from huggingface_hub import hf_hub_download
+        from safetensors.torch import load_model
+        finetune_path = hf_hub_download(repo_id="TenStrip/10Eros-Max", filename="10Eros_Max_h3_fl2va_bf16_test1.safetensors")
+        load_model(pipe.transformer, finetune_path, strict=False, device="cpu")
+        pipe.transformer.to(torch.bfloat16)
         pipe.transformer.set_attention_backend(ATTENTION)
 
         # Still startup, still free: an AoTI package carries no weights and opens its archive lazily inside the GPU