Update train_voxcpm_finetune.py

修改了issue#185中提到的问题，在训练时进行validate会对原模型执行to(torch.bfloat16)然后to(torch.float32)的操作，这样可能导致模型数值浮动，因此这个修改让validate步骤保留原模型数值

scripts/train_voxcpm_finetune.py

@@ -477,20 +477,28 @@ def generate_sample_audio(model, val_ds, audio_vae, writer, step, accelerator, s
                 log(f"[Audio] Loaded reference audio for sample {i}: duration={len(ref_audio_np)/sample_rate:.2f}s")
         except Exception as e:
             log(f"[Warning] Failed to load reference audio: {e}")
-        
+
+        # 记录原模式，避免异常时状态不一致
+        prev_training = unwrapped_model.training
         try:
             # Inference setup
             unwrapped_model.eval()
-            unwrapped_model.to(torch.bfloat16)
+            # unwrapped_model.to(torch.bfloat16)
             unwrapped_model.audio_vae = audio_vae.to(torch.float32)
             
             log(f"[Audio] Generating sample {i} with text: '{text[:50]}...'")
+            autocast_ctx = (
+                torch.autocast(device_type="cuda", dtype=torch.bfloat16)
+                if torch.cuda.is_available()
+                else contextlib.nullcontext()
+            )
             with torch.no_grad():
-                generated = unwrapped_model.generate(target_text=text, inference_timesteps=10, cfg_value=2.0)
+                with autocast_ctx:
+                    generated = unwrapped_model.generate(target_text=text, inference_timesteps=10, cfg_value=2.0)
             
             # Restore training setup
-            unwrapped_model.to(torch.float32)
+            # unwrapped_model.to(torch.float32)
-            unwrapped_model.audio_vae = None
+            # unwrapped_model.audio_vae = None
             
             if generated is None or len(generated) == 0:
                 log(f"[Warning] Generated audio is empty for sample {i}")
@@ -523,6 +531,18 @@ def generate_sample_audio(model, val_ds, audio_vae, writer, step, accelerator, s
             import traceback
             traceback.print_exc()
 
+        finally:
+            # Restore training setup（无论成功失败都恢复）
+            try:
+                # unwrapped_model.to(torch.float32)
+                unwrapped_model.audio_vae = None
+                if prev_training:
+                    unwrapped_model.train()
+                else:
+                    unwrapped_model.eval()
+            except Exception as e:
+                log(f"[Warning] Failed to restore model state: {e}")
+
 
 def load_checkpoint(model, optimizer, scheduler, save_dir: Path):
     """
@@ -671,4 +691,4 @@ if __name__ == "__main__":
     else:
         # Otherwise use command line args (parsed by argbind)
         with argbind.scope(args):
-            train()
+            train()