fix: move baseline sample after inference flag stripping

_save_sample("baseline") was called before the vocoder's inference
tensors were sanitized, causing "Inference tensors do not track version
counter". Moved it after the clone/detach loop and vocoder.to(device).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

nodes/selva_bigvgan_trainer.py

@@ -951,8 +951,6 @@ def _do_train(vocoder, mel_converter, clips,
         except Exception as e:
             print(f"[BigVGAN] Sample save failed ({label}): {e}", flush=True)
 
-    _save_sample("baseline")
-
     # Sanitize all inference tensors in the vocoder.
     # Three categories to handle (all loaded in ComfyUI's inference_mode):
     #
@@ -994,6 +992,9 @@ def _do_train(vocoder, mel_converter, clips,
     # After cloning, vocoder may be on CPU (offloaded before training).
     vocoder.to(device, dtype)
 
+    # Baseline sample — after sanitization so vocoder has normal tensors.
+    _save_sample("baseline")
+
     # ── GAFilter injection ─────────────────────────────────────────────────
     # GAFilter params are fresh tensors — no inference flag to strip.
     if use_gafilter: