fix: complete shared generator cleanup coverage

Move generator close handling into a shared utility and wire the core generation pipeline through it so partially-consumed prompt cache generators are cleaned up consistently across both model variants and the public VoxCPM wrapper.

Made-with: Cursor

src/voxcpm/core.py

@@ -8,6 +8,7 @@ from typing import Generator, Optional
 from huggingface_hub import snapshot_download
 from .model.voxcpm import VoxCPMModel, LoRAConfig
 from .model.voxcpm2 import VoxCPM2Model
+from .model.utils import next_and_close
 
 
 class VoxCPM:
@@ -171,7 +172,7 @@ class VoxCPM:
         )
 
     def generate(self, *args, **kwargs) -> np.ndarray:
-        return next(self._generate(*args, streaming=False, **kwargs))
+        return next_and_close(self._generate(*args, streaming=False, **kwargs))
 
     def generate_streaming(self, *args, **kwargs) -> Generator[np.ndarray, None, None]:
         return self._generate(*args, streaming=True, **kwargs)
@@ -292,7 +293,14 @@ class VoxCPM:
                 streaming=streaming,
             )
 
-            for wav, _, _ in generate_result:
+            if streaming:
+                try:
+                    for wav, _, _ in generate_result:
+                        yield wav.squeeze(0).cpu().numpy()
+                finally:
+                    generate_result.close()
+            else:
+                wav, _, _ = next_and_close(generate_result)
                 yield wav.squeeze(0).cpu().numpy()
 
         finally:

src/voxcpm/model/utils.py

@@ -3,6 +3,16 @@ import torch
 from transformers import PreTrainedTokenizer
 
 
+# Ref: https://github.com/OpenBMB/VoxCPM/issues/256#issuecomment-4235252732
+# Explicitly close partially-consumed generators so inference_mode cleanup
+# does not get deferred to Python's GC/finalizer path.
+def next_and_close(gen):
+    try:
+        return next(gen)
+    finally:
+        gen.close()
+
+
 def mask_multichar_chinese_tokens(tokenizer: PreTrainedTokenizer):
     """Create a tokenizer wrapper that converts multi-character Chinese tokens to single characters.
 

src/voxcpm/model/voxcpm.py

@@ -44,17 +44,7 @@ from ..modules.layers.lora import apply_lora_to_named_linear_modules
 from ..modules.locdit import CfmConfig, UnifiedCFM, VoxCPMLocDiT
 from ..modules.locenc import VoxCPMLocEnc
 from ..modules.minicpm4 import MiniCPM4Config, MiniCPMModel
-from .utils import get_dtype, mask_multichar_chinese_tokens, resolve_runtime_device
+from .utils import get_dtype, mask_multichar_chinese_tokens, next_and_close, resolve_runtime_device
-
-
-# Ref: https://github.com/OpenBMB/VoxCPM/issues/256#issuecomment-4234809321
-# Explicitly close partially-consumed generators so inference_mode cleanup
-# does not get deferred to Python's GC/finalizer path.
-def _next_and_close(gen):
-    try:
-        return next(gen)
-    finally:
-        gen.close()
 
 
 class VoxCPMEncoderConfig(BaseModel):
@@ -345,7 +335,7 @@ class VoxCPMModel(nn.Module):
         return get_dtype(self.config.dtype)
 
     def generate(self, *args, **kwargs) -> torch.Tensor:
-        return _next_and_close(self._generate(*args, streaming=False, **kwargs))
+        return next_and_close(self._generate(*args, streaming=False, **kwargs))
 
     def generate_streaming(self, *args, **kwargs) -> Generator[torch.Tensor, None, None]:
         return self._generate(*args, streaming=True, **kwargs)
@@ -471,7 +461,7 @@ class VoxCPMModel(nn.Module):
                     yield decode_audio
                 break
             else:
-                latent_pred, pred_audio_feat = _next_and_close(inference_result)
+                latent_pred, pred_audio_feat = next_and_close(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
                         print(
@@ -579,7 +569,7 @@ class VoxCPMModel(nn.Module):
         return merged_cache
 
     def generate_with_prompt_cache(self, *args, **kwargs) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
-        return _next_and_close(self._generate_with_prompt_cache(*args, streaming=False, **kwargs))
+        return next_and_close(self._generate_with_prompt_cache(*args, streaming=False, **kwargs))
 
     def generate_with_prompt_cache_streaming(
         self, *args, **kwargs
@@ -698,7 +688,7 @@ class VoxCPMModel(nn.Module):
                     yield (decode_audio, target_text_token, pred_audio_feat)
                 break
             else:
-                latent_pred, pred_audio_feat = _next_and_close(inference_result)
+                latent_pred, pred_audio_feat = next_and_close(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
                         print(
@@ -721,7 +711,7 @@ class VoxCPMModel(nn.Module):
             yield (decode_audio, target_text_token, pred_audio_feat)
 
     def inference(self, *args, **kwargs) -> Tuple[torch.Tensor, torch.Tensor]:
-        return _next_and_close(self._inference(*args, streaming=False, **kwargs))
+        return next_and_close(self._inference(*args, streaming=False, **kwargs))
 
     def inference_streaming(self, *args, **kwargs) -> Generator[Tuple[torch.Tensor, List[torch.Tensor]], None, None]:
         return self._inference(*args, streaming=True, **kwargs)

src/voxcpm/model/voxcpm2.py

@@ -45,17 +45,7 @@ from ..modules.layers.lora import apply_lora_to_named_linear_modules
 from ..modules.locdit import CfmConfig, UnifiedCFM, VoxCPMLocDiTV2
 from ..modules.locenc import VoxCPMLocEnc
 from ..modules.minicpm4 import MiniCPM4Config, MiniCPMModel
-from .utils import get_dtype, mask_multichar_chinese_tokens, resolve_runtime_device
+from .utils import get_dtype, mask_multichar_chinese_tokens, next_and_close, resolve_runtime_device
-
-
-# Ref: https://github.com/OpenBMB/VoxCPM/issues/256#issuecomment-4234809321
-# Explicitly close partially-consumed generators so inference_mode cleanup
-# does not get deferred to Python's GC/finalizer path.
-def _next_and_close(gen):
-    try:
-        return next(gen)
-    finally:
-        gen.close()
 
 
 # A simple function to trim audio silence using VAD, not used default
@@ -451,7 +441,7 @@ class VoxCPM2Model(nn.Module):
         return tokens, feats, t_mask, a_mask
 
     def generate(self, *args, **kwargs) -> torch.Tensor:
-        return _next_and_close(self._generate(*args, streaming=False, **kwargs))
+        return next_and_close(self._generate(*args, streaming=False, **kwargs))
 
     def generate_streaming(self, *args, **kwargs) -> Generator[torch.Tensor, None, None]:
         return self._generate(*args, streaming=True, **kwargs)
@@ -651,7 +641,7 @@ class VoxCPM2Model(nn.Module):
                     yield decode_audio
                 break
             else:
-                latent_pred, pred_audio_feat, context_len = _next_and_close(inference_result)
+                latent_pred, pred_audio_feat, context_len = next_and_close(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
                         print(
@@ -769,7 +759,7 @@ class VoxCPM2Model(nn.Module):
         return merged
 
     def generate_with_prompt_cache(self, *args, **kwargs) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
-        return _next_and_close(self._generate_with_prompt_cache(*args, streaming=False, **kwargs))
+        return next_and_close(self._generate_with_prompt_cache(*args, streaming=False, **kwargs))
 
     def generate_with_prompt_cache_streaming(
         self, *args, **kwargs
@@ -938,7 +928,7 @@ class VoxCPM2Model(nn.Module):
                     yield (decode_audio, target_text_token, pred_audio_feat)
                 break
             else:
-                latent_pred, pred_audio_feat, context_len = _next_and_close(inference_result)
+                latent_pred, pred_audio_feat, context_len = next_and_close(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
                         print(
@@ -961,7 +951,7 @@ class VoxCPM2Model(nn.Module):
             yield (decode_audio, target_text_token, pred_audio_feat)
 
     def inference(self, *args, **kwargs) -> Tuple[torch.Tensor, torch.Tensor]:
-        feat_pred, generated_feat, _ = _next_and_close(self._inference(*args, streaming=False, **kwargs))
+        feat_pred, generated_feat, _ = next_and_close(self._inference(*args, streaming=False, **kwargs))
         return feat_pred, generated_feat
 
     def inference_streaming(self, *args, **kwargs) -> Generator[Tuple[torch.Tensor, List[torch.Tensor]], None, None]: