fix routes problem and update

Author: rich7420

.pre-commit-config.yaml

@@ -71,7 +71,7 @@ repos:
         name: cargo clippy
         language: system
         pass_filenames: false
-        entry: bash -c 'LIBTORCH_USE_PYTORCH=1 LIBTORCH_BYPASS_VERSION_CHECK=1 cargo clippy "$@"' --
+        entry: bash -c 'LIBTORCH_USE_PYTORCH=1 cargo clippy "$@"' --
         args:
           - --manifest-path
           - qdp/Cargo.toml

qdp/qdp-python/benchmark/benchmark_latency.py

@@ -18,28 +18,20 @@
 """
 Data-to-State latency benchmark: CPU RAM -> GPU VRAM.
 
-Run:
-    python qdp/qdp-python/benchmark/benchmark_latency.py --qubits 16 \
+Run from qdp-python directory (qumat_qdp must be importable, e.g. via uv):
+    uv run python benchmark/benchmark_latency.py --qubits 16 \\
         --batches 200 --batch-size 64 --prefetch 16
 """
 
 from __future__ import annotations
 
 import argparse
-import sys
 import time
-from pathlib import Path
 
 import torch
 
-# Add project root to path so qumat_qdp is importable when run as script
-_script_dir = Path(__file__).resolve().parent
-_project_root = _script_dir.parent
-if str(_project_root) not in sys.path:
-    sys.path.insert(0, str(_project_root))
-
-from benchmark.utils import normalize_batch, prefetched_batches  # noqa: E402
-from qumat_qdp import QdpBenchmark  # noqa: E402
+from benchmark.utils import normalize_batch, prefetched_batches
+from qumat_qdp import QdpBenchmark
 
 BAR = "=" * 70
 SEP = "-" * 70

qdp/qdp-python/benchmark/benchmark_loader_throughput.py

@@ -21,24 +21,16 @@
 Compares iterator-based throughput with run_throughput_pipeline_py.
 Expectation: loader version slightly slower due to Python boundary per batch.
 
-Usage (from qdp-python):
+Run from qdp-python directory (qumat_qdp must be importable, e.g. via uv):
   uv run python benchmark/benchmark_loader_throughput.py --qubits 16 --batches 200 --batch-size 64
 """
 
 from __future__ import annotations
 
 import argparse
-import sys
 import time
-from pathlib import Path
 
-# Add project root to path so qumat_qdp is importable when run as script
-_script_dir = Path(__file__).resolve().parent
-_project_root = _script_dir.parent
-if str(_project_root) not in sys.path:
-    sys.path.insert(0, str(_project_root))
-
-from qumat_qdp import QuantumDataLoader, QdpBenchmark  # noqa: E402
+from qumat_qdp import QuantumDataLoader, QdpBenchmark
 
 
 def run_loader_throughput(

qdp/qdp-python/benchmark/benchmark_throughput.py

@@ -23,26 +23,18 @@
 - Prefetch on the CPU side to keep the GPU fed.
 - Encode vectors into amplitude states on GPU and run a tiny consumer op.
 
-Run:
-    python qdp/benchmark/benchmark_throughput.py --qubits 16 --batches 200 --batch-size 64
+Run from qdp-python directory (qumat_qdp must be importable, e.g. via uv):
+    uv run python benchmark/benchmark_throughput.py --qubits 16 --batches 200 --batch-size 64
 """
 
 import argparse
-import sys
 import time
-from pathlib import Path
 
 import numpy as np
 import torch
 
-# Add project root to path so qumat_qdp is importable when run as script
-_script_dir = Path(__file__).resolve().parent
-_project_root = _script_dir.parent
-if str(_project_root) not in sys.path:
-    sys.path.insert(0, str(_project_root))
-
-from benchmark.utils import normalize_batch, prefetched_batches  # noqa: E402
-from qumat_qdp import QdpBenchmark  # noqa: E402
+from benchmark.utils import normalize_batch, prefetched_batches
+from qumat_qdp import QdpBenchmark
 
 BAR = "=" * 70
 SEP = "-" * 70

qdp/qdp-python/qumat_qdp/__init__.py

@@ -29,7 +29,8 @@
 
 from __future__ import annotations
 
-# Rust extension (built by maturin)
+# Rust extension (built by maturin). QdpEngine/QuantumTensor are public for
+# advanced use; QdpBenchmark and QuantumDataLoader are the recommended high-level API.
 import _qdp
 
 from qumat_qdp.api import (

qdp/qdp-python/qumat_qdp/loader.py

@@ -31,7 +31,8 @@
 
 from typing import Iterator, Optional
 
-# Lazy import _qdp until __iter__ is used
+# Lazy import _qdp until __iter__ is used: avoids import-time load when only
+# importing QuantumDataLoader, and defers extension/CUDA requirement until first use.
 _qdp: Optional[object] = None