Fix ChatTTS CUDA device-side assert with text sanitize and GPU recovery.

Re-enable KV cache by default, normalize digits and unsafe chars, disable per-chunk split_text, and reload ChatTTS after CUDA errors.

Co-authored-by: Cursor <cursoragent@cursor.com>

config.py

@@ -39,6 +39,13 @@ def _env_int(key: str, default: int) -> int:
         return default
 
 
+def _env_bool(key: str, default: bool) -> bool:
+    raw = os.environ.get(key)
+    if raw is None:
+        return default
+    return raw.strip().lower() in ("1", "true", "yes", "on")
+
+
 # ---------------------------------------------------------------------------
 # 网络与服务
 # ---------------------------------------------------------------------------
@@ -145,6 +152,9 @@ TTS_MAX_NEW_TOKEN = _env_int("TTS_MAX_NEW_TOKEN", 1024)
 # 至少生成多少 audio token 才允许结束（防止首 token EOS → 无限递归重试）
 TTS_MIN_NEW_TOKEN = _env_int("TTS_MIN_NEW_TOKEN", 16)
 
+# GPT KV cache（关闭可省显存，但部分 transformers 版本会触发 CUDA assert）
+TTS_ENABLE_CACHE = _env_bool("TTS_ENABLE_CACHE", True)
+
 # ---------------------------------------------------------------------------
 # 上传临时文件目录
 # ---------------------------------------------------------------------------