Error that kill comfyUI server (segmentation fault)

[FR-Mister-T]

hello,

it's my first try and i'm running into an error that kill comfyui server
"Segmentation fault (core dumped)"

Unfortunately the system does not give more details and the core dump is like 400mb file :(
i'm running on venv python 3.13 / Ubuntu with pytorch ROCm 7.1.1 and the Q8 Vibevoice large

I'll try an other weight to see if it does the same

[VibeVoice] Loading model from: /home/zeuss194/COMFY/ComfyUI/models/vibevoice/VibeVoiceLargeQ8
[VibeVoice] Loading 8bit quantized model...
Loading checkpoint shards: 100%|██████████████████| 3/3 [00:06<00:00, 2.20s/it]
[VibeVoice] Model loaded in 7.36 seconds
[VibeVoice] Loading VibeVoice processor...
[VibeVoice] Found Qwen tokenizer in: /home/zeuss194/COMFY/ComfyUI/models/vibevoice/tokenizer
[VibeVoice] Found complete tokenizer at: /home/zeuss194/COMFY/ComfyUI/models/vibevoice/tokenizer
[VibeVoice] Quantized model already mapped to device via device_map
[VibeVoice] Processing text segment 1 (10 words)
[VibeVoice] Starting audio generation with 20 diffusion steps...
[VibeVoice] Generating audio with 20 diffusion steps...
[VibeVoice] Note: Progress bar shows max possible tokens, not actual needed (~30 estimated)
[VibeVoice] The generation will stop automatically when audio is complete
Generating (active: 1/1): 0%| | 0/462 [00:00<?, ?it/s] Segmentation fault (core dumped)

[EDIT] 1.5B Microsoft weight works fine
[EDIT 2] Large model weight works fine
This is most likely related to how quantized models are handled on AMD hardware (or not xD)