Fix/14 gpu round trip

[LucaLM02]

Summary

Keeps data on the GPU on the device="gpu" path (closes #14 ) :

• Truncation SVD now runs on the array's own device (xp.linalg.svd) instead of copying R to the host with .get() and calling np.linalg.svd.
• New to_host: bool = True argument on apply / compress. With to_host=False the result stays backed by CuPy arrays and can be fed straight into the next call without a device/host round trip. The default preserves the previous behaviour, so existing callers are unaffected.
• CPU path is unchanged.

Performance

Measured on Google Colab:

• GPU: Tesla T4 (15 GB), driver 580.82.07
• CPU: Intel Xeon @ 2.00 GHz, 2 vCPU
• cupy 14.2.0, CUDA runtime 12.9, numpy 2.4.4, Python 3.14.8
• before = fe1779b (main), after = fbec6ae (this branch)

Method: complex128, seed=0, random MPO × MPS, 20 sites, phys_dim=4, median of 10 runs. Each path is warmed up first, and before/after are timed in separate processes on the same session.

chi_out = 64 (input bond dim 64)

Scenario	before (main)	after, to_host=True	after, to_host=False	speedup
single apply	613.5 ms	614.2 ms	610.0 ms	1.01x
5 chained apply	3016.6 ms	3018.3 ms	2978.2 ms	1.01x

chi_out = 128 (input bond dim 128, after only)

Scenario	to_host=True	to_host=False
single apply	8455.1 ms	8448.3 ms
3 chained apply	24863.1 ms	24830.8 ms

Testing

• Full test suite on the T4: 40 passed in ~70 s.
• Chained apply(..., to_host=False) calls accept CuPy inputs and run without host round trips of the result.

Acceptance criteria

• No .get() / D2H transfer inside the per-site loop
• Chained GPU apply calls do not round-trip the result through host memory
• CPU results and numerics unchanged
• GPU tests pass
• Before/after timings recorded above, with hardware

[Copilot]

Copilot review overview

🟡 Changes recommended

The two-site fallback still round-trips GPU data through the host, and the GPU SVD path lacks direct coverage.

Review effort: Balanced
Findings: 1 · 2

Open (3)

• Keep two-site fallback SVD backend-aware on GPU · New
• Remove duplicated period in return description · New
• Add GPU test covering positive-cutoff device-local SVD · New

What changed in this PR

Moves GPU SVD work and optional outputs fully onto the selected device.

Changes:

• Adds to_host to public compression APIs.
• Runs truncated SVD through the active array backend.
• Adds GPU output and chaining tests.

File	Description
src/​src_method/​apply.py	Exposes to_host for application results.
src/​src_method/​compress.py	Exposes to_host for compression results.
src/​src_method/​stack.py	Propagates output-device selection.
src/​src_method/​_sweep.py	Avoids unconditional host conversion.
src/​src_method/​utils/​linalg.py	Runs truncation SVD on the active backend.
tests/​test_gpu_backend.py	Tests GPU-backed outputs and chaining.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.