Faster-Whisper-XXL Pro Error when using custom model

When I use Faster-Whisper-XXL Pro r3.256.1 with the model [SoybeanMilk/faster-whisper-Breeze-ASR-25](https://huggingface.co/SoybeanMilk/faster-whisper-Breeze-ASR-25) the program crashes during initialization with the following error: `Exception: data did not match any variant of untagged enum ModelWrapper at line 264874 column 3` If I replace this model’s tokenizer.json from another working model (for example, a turbo model), the program can start and run, but of course the transcription result not what I expect. I am wondering if this is related to the faster-whisper version used in this standalone build. Environment Ubuntu 22.04.4 LTS Standalone Faster-Whisper-XXL Pro r3.256.1 Backend: CUDA GPU: NVIDIA GeForce RTX 3090 Model: [SoybeanMilk/faster-whisper-Breeze-ASR-25](https://huggingface.co/SoybeanMilk/faster-whisper-Breeze-ASR-25) Log excerpt: ``` Standalone Faster-Whisper-XXL Pro r3.256.1 running on: CUDA -= Licensed to: xxxxxxxxxxxxxxx =- Number of visible GPU devices: 1 Supported compute types by GPU: {'float16', 'float32', 'int8_bfloat16', 'int8_float32', 'int8', 'int8_float16', 'bfloat16'} [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] CPU: GenuineIntel (SSE4.1=true, AVX=true, AVX2=true, AVX512=false) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - Selected ISA: AVX2 [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - Use Intel MKL: true [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - SGEMM backend: MKL (packed: false) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - GEMM_S16 backend: MKL (packed: false) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - GEMM_S8 backend: MKL (packed: false, u8s8 preferred: true) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] GPU #0: NVIDIA GeForce RTX 3090 (CC=8.6) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - Allow INT8: true [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - Allow FP16: true (with Tensor Cores: true) [2025-12-03 11:31:26.532] [ctranslate2] [thread 493287] [info] - Allow BF16: true [2025-12-03 11:31:27.455] [ctranslate2] [thread 493287] [info] Using CUDA allocator: cuda_malloc_async [2025-12-03 11:31:27.715] [ctranslate2] [thread 493287] [info] Loaded model /srv/vghks/STC/whiteblocktechSTC/_models/faster-whisper-Breeze-ASR-25 on device cuda:0 [2025-12-03 11:31:27.715] [ctranslate2] [thread 493287] [info] - Binary version: 6 [2025-12-03 11:31:27.715] [ctranslate2] [thread 493287] [info] - Model specification revision: 3 [2025-12-03 11:31:27.715] [ctranslate2] [thread 493287] [info] - Selected compute type: float16 Traceback (most recent call last): File "__main__.py", line 2410, in <module> File "__main__.py", line 2069, in cli File "faster_whisper/transcribe.py", line 830, in __init__ Exception: data did not match any variant of untagged enum ModelWrapper at line 264874 column 3 [PYI-493287:ERROR] Failed to execute script '__main__' due to unhandled exception! ``` To check whether this is a model issue, I also tested the same model with the official faster-whisper Python package in a clean environment. In this setup, the model loads successfully and transcription works fine.