Remove CUDA-specific early exit from XPU test

[Silv3S]

Fixes #2509

From amp docs:

torch.cuda.amp.autocast(args...) and torch.cpu.amp.autocast(args...) is deprecated.
Please use torch.amp.autocast("cuda", args...) or torch.amp.autocast("cpu", args...) instead.

This PR fixes test_pickle_gradscaler_xpu, as it tries to use torch.xpu.amp that doesn't exist and also calls CUDA-specific early exit. We should just drop this check and proceed with the test itself (it passes).

Second test from issue doesn't require any work, as test_grad_scaler_deprecated_warning_xpu checks if deprecation warning is raised but it doesn't make sense for XPU, as XPU doesn't have that deprecated torch.xpu.amp implemented

[Copilot]

Pull request overview

Removes an XPU-specific assertion path in the GradScaler pickling test that referenced torch.xpu.amp (which doesn’t exist), aligning the XPU test behavior with the non-CUDA path and preventing AttributeError failures.

Changes:

• Drop the XPU-only torch.xpu.amp.common.amp_definitely_not_available() check in test_pickle_gradscaler.
• Let XPU follow the existing non-CUDA assertion path (a.is_enabled() expected true).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.