Upload app.py with huggingface_hub

app.py

@@ -82,35 +82,24 @@ if patch.exists():
 ip_file = SAM3D_PATH / "sam3d_objects" / "pipeline" / "inference_pipeline.py"
 if ip_file.exists():
     ip_src = ip_file.read_text()
-    # Replace the set_attention_backend function to respect our env vars
-    old_fn = """def set_attention_backend():
-    if torch.cuda.is_available():
-        gpu_name = torch.cuda.get_device_name(0)
-    else:
-        gpu_name = "CPU"
-
-    logger.info(f"GPU name is {gpu_name}")
-    if "A100" in gpu_name or "H100" in gpu_name or "H200" in gpu_name:
-        # logger.info("Use flash_attn")
-        os.environ["ATTN_BACKEND"] = "flash_attn"
-        os.environ["SPARSE_ATTN_BACKEND"] = "flash_attn""""
-    new_fn = """def set_attention_backend():
-    if torch.cuda.is_available():
-        gpu_name = torch.cuda.get_device_name(0)
-    else:
-        gpu_name = "CPU"
-
-    logger.info(f"GPU name is {gpu_name}")
-    # PATCHED: Always use sdpa backend (flash_attn not available on ZeroGPU)
-    logger.info("Using sdpa backend (patched for ZeroGPU)")
-    os.environ.setdefault("ATTN_BACKEND", "sdpa")
-    os.environ.setdefault("SPARSE_ATTN_BACKEND", "sdpa")""""
-    if old_fn in ip_src:
-        ip_src = ip_src.replace(old_fn, new_fn)
+    # Find and replace the set_attention_backend function
+    old_marker = 'os.environ["ATTN_BACKEND"] = "flash_attn"'
+    if old_marker in ip_src:
+        # Replace the entire if-block that forces flash_attn
+        ip_src = ip_src.replace(
+            'if "A100" in gpu_name or "H100" in gpu_name or "H200" in gpu_name:\n'
+            '        # logger.info("Use flash_attn")\n'
+            '        os.environ["ATTN_BACKEND"] = "flash_attn"\n'
+            '        os.environ["SPARSE_ATTN_BACKEND"] = "flash_attn"',
+            '# PATCHED: Always use sdpa backend (flash_attn not available on ZeroGPU)\n'
+            '    logger.info("Using sdpa backend (patched for ZeroGPU)")\n'
+            '    os.environ.setdefault("ATTN_BACKEND", "sdpa")\n'
+            '    os.environ.setdefault("SPARSE_ATTN_BACKEND", "sdpa")'
+        )
         ip_file.write_text(ip_src)
         print("PATCHED: inference_pipeline.py - forced sdpa backend")
     else:
-        print("WARNING: Could not patch inference_pipeline.py")
+        print("INFO: inference_pipeline.py already patched or different version")
 
 sys.path.insert(0, str(SAM3D_PATH))
 sys.path.insert(0, str(SAM3D_PATH / "notebook"))