RXT 5070 Ti - not working multilanguage

[anielka41]

Hi,
(chatterbox) tomasz@tomasz-MS-7D18:~/TTS/Chatterbox-TTS-Server$ ./start.sh

============================================================
Chatterbox TTS Server - Launcher

Checking for Python installation...

[OK] Found Python 3.10.19 (python3)

Verifying Python version...
[OK] Python 3.10 meets requirements (3.10+)

Checking for Python venv module...
[OK] Python venv module available

============================================================
Starting Chatterbox TTS Server...

Using: python3

============================================================
Chatterbox TTS Server - Launcher

[1/6] Checking Python installation...
✓ Python 3.10.19 detected

[2/6] Setting up environment...
→ Project directory: /home/tomasz/TTS/Chatterbox-TTS-Server
→ Virtual environment: /home/tomasz/TTS/Chatterbox-TTS-Server/venv

[3/6] Checking existing installation...
✓ Found existing NVIDIA GPU (CUDA 12.8 / Blackwell) installation

[4/6] Using existing installation...
✓ Installation type: NVIDIA GPU (CUDA 12.8 / Blackwell)

[5/6] Loading configuration...
✓ Server will run on 0.0.0.0:8004

[6/6] Launching Chatterbox TTS Server...
→ Starting server.py...
→ Waiting for server to start (this may take a few minutes on first run)...
.2026-02-18 10:46:10 [INFO] main: Starting TTS Server directly on http://0.0.0.0:8004
2026-02-18 10:46:10 [INFO] main: API documentation will be available at http://0.0.0.0:8004/docs
2026-02-18 10:46:10 [INFO] main: Web UI will be available at http://0.0.0.0:8004/
INFO: Started server process [13418]
INFO: Waiting for application startup.
2026-02-18 10:46:10 [INFO] server: TTS Server: Initializing application...
2026-02-18 10:46:10 [INFO] server: Configuration loaded. Log file at: /home/tomasz/TTS/Chatterbox-TTS-Server/logs/tts_server.log
/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/cuda/init.py:235: UserWarning:
NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

warnings.warn(
2026-02-18 10:46:10 [INFO] engine: CUDA requested and functional. Using CUDA.
2026-02-18 10:46:10 [INFO] engine: Final device selection: cuda
2026-02-18 10:46:10 [INFO] engine: Model selector from config: 'chatterbox-multilingual'
2026-02-18 10:46:10 [INFO] engine: Model selector 'chatterbox-multilingual' resolved to Multilingual model (ChatterboxMultilingualTTS)
2026-02-18 10:46:10 [INFO] engine: Initializing ChatterboxMultilingualTTS on device 'cuda'...
2026-02-18 10:46:10 [INFO] engine: Model type: multilingual
Fetching 6 files: 100%|████████████████████████| 6/6 [00:00<00:00, 18808.54it/s]
2026-02-18 10:46:11 [ERROR] engine: Failed to load model using from_pretrained: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
Traceback (most recent call last):
File "/home/tomasz/TTS/Chatterbox-TTS-Server/engine.py", line 330, in load_model
chatterbox_model = model_class.from_pretrained(device=model_device)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/chatterbox/mtl_tts.py", line 212, in from_pretrained
return cls.from_local(ckpt_dir, device)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/chatterbox/mtl_tts.py", line 176, in from_local
ve.to(device).eval()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1343, in to
return self._apply(convert)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 903, in _apply
module._apply(fn)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 290, in _apply
self._init_flat_weights()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 215, in _init_flat_weights
self.flatten_parameters()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 271, in flatten_parameters
torch._cudnn_rnn_flatten_weight(
RuntimeError: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

2026-02-18 10:46:11 [CRITICAL] server: CRITICAL: TTS Model failed to load on startup. Server might not function correctly.
2026-02-18 10:46:11 [INFO] server: Application startup sequence complete.
INFO: Application startup complete.
INFO: Uvicorn running on http://0.0.0.0:8004 (Press CTRL+C to quit)

  ✓ Server ready! (took 4.0s)

2026-02-18 10:50:19 [INFO] server: Request received for main UI page ('/').
2026-02-18 10:50:19 [INFO] server: Request received for /api/ui/initial-data.
2026-02-18 10:50:19 [INFO] utils: Found 28 predefined voices in /home/tomasz/TTS/Chatterbox-TTS-Server/voices
2026-02-18 10:50:25 [INFO] server: Request received for /save_settings.
2026-02-18 10:50:25 [INFO] config: TTS processing device resolved to: cuda
2026-02-18 10:50:25 [INFO] config: Configuration successfully saved to config.yaml
2026-02-18 10:50:25 [INFO] config: Configuration updated, saved, and re-resolved successfully.
2026-02-18 10:50:25 [INFO] server: Request received for /restart_server (Model Hot-Swap).
2026-02-18 10:50:25 [INFO] engine: Initiating model hot-swap/reload sequence...
2026-02-18 10:50:25 [INFO] engine: Python garbage collection completed.
2026-02-18 10:50:25 [INFO] engine: Clearing CUDA cache...
2026-02-18 10:50:25 [INFO] engine: Memory cleared. Reloading model from updated config...
2026-02-18 10:50:25 [INFO] engine: CUDA requested and functional. Using CUDA.
2026-02-18 10:50:25 [INFO] engine: Final device selection: cuda
2026-02-18 10:50:25 [INFO] engine: Model selector from config: 'chatterbox-multilingual'
2026-02-18 10:50:25 [INFO] engine: Model selector 'chatterbox-multilingual' resolved to Multilingual model (ChatterboxMultilingualTTS)
2026-02-18 10:50:25 [INFO] engine: Initializing ChatterboxMultilingualTTS on device 'cuda'...
2026-02-18 10:50:25 [INFO] engine: Model type: multilingual
Fetching 6 files: 100%|████████████████████████| 6/6 [00:00<00:00, 18027.09it/s]
2026-02-18 10:50:26 [ERROR] engine: Failed to load model using from_pretrained: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
Traceback (most recent call last):
File "/home/tomasz/TTS/Chatterbox-TTS-Server/engine.py", line 330, in load_model
chatterbox_model = model_class.from_pretrained(device=model_device)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/chatterbox/mtl_tts.py", line 212, in from_pretrained
return cls.from_local(ckpt_dir, device)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/chatterbox/mtl_tts.py", line 176, in from_local
ve.to(device).eval()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1343, in to
return self._apply(convert)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 903, in _apply
module._apply(fn)
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 290, in _apply
self._init_flat_weights()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 215, in _init_flat_weights
self.flatten_parameters()
File "/home/tomasz/TTS/Chatterbox-TTS-Server/venv/lib/python3.10/site-packages/torch/nn/modules/rnn.py", line 271, in flatten_parameters
torch._cudnn_rnn_flatten_weight(
RuntimeError: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

2026-02-18 10:50:26 [ERROR] server: Model reload failed. The server may be in an inconsistent state. Check logs for details.

[htrex]

Hi @anielka41,

This is not a multilingual model issue — it's a PyTorch / CUDA compatibility problem with your GPU. The same error would happen with the Original or Turbo model.

Your RTX 5070 Ti uses the Blackwell architecture (sm_120), but the installed PyTorch only supports up to sm_90, so any model will fail when loading to GPU.

Try a clean reinstall with the correct CUDA 12.8 flag for Blackwell GPUs:

python3 start.py --reinstall --nvidia-cu128

This should fix it for all models (Original, Turbo, and Multilingual).

[devnen]

The underlying issue here is the RTX 5070 Ti needing CUDA 12.8 / sm_120 support. This has been fixed in commit b181edb. Please reinstall with the Blackwell option:

python start.py --reinstall --nvidia-cu128

This should fix GPU detection for all models including multilingual.