Can we use --quantization None to disable quantization? #632
Description
Describe the bug
Can we use --quantization None to disable quantization?
Desktop (please complete the following information):
- OS: [e.g. iOS] Win10
- Browser [e.g. chrome, safari] e.g.
- Version [e.g. 22] (c11a74e)
Additional context
ACE-Step 1.5 - AMD ROCm 7.2 Edition
[Info] Using embedded Python at D:\ACE-Step-1.5\python_embeded
GPU: AMD Radeon RX 7900 XT
HIP: 7.2.0
Starting ACE-Step Gradio Web UI...
Server will be available at: http://127.0.0.1:7860
Loaded configuration from D:\ACE-Step-1.5.env
2026-02-18 15:00:01.308 | WARNING | acestep.training.trainer::36 - bitsandbytes not installed. Using standard AdamW.
2026-02-18 15:00:05.003 | INFO | acestep.gpu_config:get_gpu_memory_gb:388 - ROCm GPU detected: AMD Radeon RX 7900 XT (20.0 GB, HIP 7.2.0)
============================================================
GPU Configuration Detected:
GPU Memory: 19.98 GB
Configuration Tier: tier6a
Max Duration (with LM): 480s (8 min)
Max Duration (without LM): 600s (10 min)
Max Batch Size (with LM): 4
Max Batch Size (without LM): 8
Default LM Init: True
Available LM Models: ['acestep-5Hz-lm-0.6B', 'acestep-5Hz-lm-1.7B']
Auto-enabling CPU offload (GPU 20.0GB < 20.0GB threshold)
Output directory: D:/ACE-Step-1.5/gradio_outputs
usage: acestep_v15_pipeline.py [-h] [--port PORT] [--share] [--debug] [--server-name SERVER_NAME]
[--language {en,zh,he,ja}] [--allowed-path ALLOWED_PATH] [--service_mode SERVICE_MODE]
[--init_service INIT_SERVICE] [--checkpoint CHECKPOINT] [--config_path CONFIG_PATH]
[--device {auto,cuda,mps,xpu,cpu}] [--init_llm INIT_LLM]
[--lm_model_path LM_MODEL_PATH] [--backend {vllm,pt,mlx}]
[--use_flash_attention USE_FLASH_ATTENTION] [--offload_to_cpu OFFLOAD_TO_CPU]
[--offload_dit_to_cpu OFFLOAD_DIT_TO_CPU]
[--quantization {int8_weight_only,int4_weight_only,None}]
[--download-source {huggingface,modelscope,auto}] [--batch_size BATCH_SIZE]
[--enable-api] [--auth-username AUTH_USERNAME] [--auth-password AUTH_PASSWORD]
[--api-key API_KEY]
acestep_v15_pipeline.py: error: argument --quantization: invalid choice: 'None' (choose from int8_weight_only, int4_weight_only, None)
请按任意键继续. . .