Upload app.py with huggingface_hub

app.py

@@ -291,6 +291,15 @@ def test_sam3d_only(image: np.ndarray):
         print(f"  Mask created: {mask.sum()} pixels ({time.time()-t0:.0f}s)")
 
         from inference import Inference
+        # SAM3D's inference_pipeline.py auto-detects H200 and sets ATTN_BACKEND=flash_attn
+        # We must override BACK to sdpa since flash_attn is not available
+        import sam3d_objects.model.backbone.tdfy_dit.modules.attention as _attn_mod
+        import sam3d_objects.model.backbone.tdfy_dit.modules.sparse as _sparse_mod
+        _attn_mod.BACKEND = "sdpa"
+        _sparse_mod.ATTN = "sdpa"
+        os.environ["ATTN_BACKEND"] = "sdpa"
+        os.environ["SPARSE_ATTN_BACKEND"] = "sdpa"
+        print(f"  Attention backends forced to sdpa")
         print(f"  Loading SAM3D... VRAM: {torch.cuda.memory_allocated()/1e9:.1f}GB")
         sam3d = Inference(CONFIG_PATH, compile=False)
         print(f"  SAM3D loaded ({time.time()-t0:.0f}s, VRAM: {torch.cuda.memory_allocated()/1e9:.1f}GB)")