Add compatibility check for torch.xpu before accessing

[onuryuruten]

Description

Adds a safety check before accessing torch.xpu.is_available() to prevent AttributeError when using PyTorch versions < 2.4.

Problem

The code currently calls torch.xpu.is_available() without checking if the xpu attribute exists. The torch.xpu module was added in PyTorch 2.4+, causing an AttributeError on earlier versions, e.g. 2.2.2:

AttributeError: module 'torch' has no attribute 'xpu'

This can break the applications running PyTorch 2.0-2.3, even though F5-TTS's pyproject.toml only requires torch>=2.0.0.

Solution

Add hasattr(torch, 'xpu') check before calling torch.xpu.is_available(). This ensures backward compatibility with PyTorch 2.0+ while still supporting Intel XPU acceleration when available.

Changes

• Modified src/f5_tts/infer/utils_infer.py line 42
• Changed: if torch.xpu.is_available():
• To: if hasattr(torch, 'xpu') and torch.xpu.is_available():

Testing

• Tested with PyTorch 2.2.2 (previously failing) ✓
• Tested with PyTorch 2.5+ (xpu available) ✓
• Application now runs successfully on both versions

Impact

• Fixes compatibility with PyTorch 2.0-2.3
• No breaking changes for existing users
• Maintains Intel XPU support for newer PyTorch versions