https://github.com/ace-step/ACE-Step-1.5 (v0.1.3, 2026-02-28)
The most powerful local music generation model that outperforms most commercial alternatives, supporting Mac, AMD, Intel, and CUDA devices.
ACE-Step 1.5 설치
# 1. Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh # macOS / Linux
# powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" # Windows
# Windows
PS C:\Developer\Sources> irm https://astral.sh/uv/install.ps1 | iex
Downloading uv 0.10.7 (x86_64-pc-windows-msvc)
Installing to C:\Users\bellworld\.local\bin
uv.exe
uvx.exe
uvw.exe
everything's installed!
# 2. Clone & install
git clone https://github.com/ACE-Step/ACE-Step-1.5.git
cd ACE-Step-1.5
uv sync
PS D:\07.DEVELOPER> cd D:\07.DEVELOPER
PS D:\07.DEVELOPER>
PS D:\07.DEVELOPER> git clone https://github.com/ACE-Step/ACE-Step-1.5.git
Cloning into 'ACE-Step-1.5'...
remote: Enumerating objects: 7428, done.
remote: Counting objects: 100% (2408/2408), done.
remote: Compressing objects: 100% (594/594), done.
remote: Total 7428 (delta 2009), reused 1815 (delta 1814), pack-reused 5020 (from 4)
Receiving objects: 100% (7428/7428), 10.52 MiB | 4.16 MiB/s, done.
Resolving deltas: 100% (4490/4490), done.
PS D:\07.DEVELOPER>
PS D:\07.DEVELOPER> cd ACE-Step-1.5
PS D:\07.DEVELOPER\ACE-Step-1.5>
PS D:\07.DEVELOPER\ACE-Step-1.5> uv sync
Using CPython 3.11.9
Creating virtual environment at: .venv
⠧ Resolving dependencies...
flash-attn ------------------------------ 19.59 MiB/239.15 MiB
# 3. Launch Gradio UI (models auto-download on first run)
uv run acestep
PS D:\07.DEVELOPER\ACE-Step-1.5> uv run acestep
Uninstalled 1 package in 79ms
░░░░░░░░░░░░░░░░░░░░ [0/1] Installing wheels... warning: Failed to hardlink files; falling back to full copy. This may lead to degraded performance.
If the cache and target directories are on different filesystems, hardlinking may not be supported.
If this is intentional, set `export UV_LINK_MODE=copy` or use `--link-mode=copy` to suppress this warning.
Installed 1 package in 389ms
Loaded configuration from D:\07.DEVELOPER\ACE-Step-1.5\.env.example (fallback)
# Or launch REST API server
uv run acestep-api
웹 페이지 접속 후 초기 설정 진행
- Running on local URL: http://127.0.0.1:7860
- To create a public link, set
share=Trueinlaunch().
초기 설정을 진행하면 해당 설정에 맞는 모델들을 다운로드 받는다.
silence_latent.pt: 100%|█████████████████████████████████████████████████████████████| 3.84M/3.84M [00:10<00:00, 351kB/s]
tokenizer.json: 100%|████████████████████████████████████████████████████████████████| 11.4M/11.4M [00:14<00:00, 780kB/s]
tokenizer_config.json: 100%|█████████████████████████████████████████████████████████| 14.1M/14.1M [00:32<00:00, 439kB/s]
tokenizer.json: 100%|████████████████████████████████████████████████████████████████| 24.3M/24.3M [00:44<00:00, 545kB/s]
model.safetensors: 2%|█ | 21.0M/1.19G [00:43<39:02, 500kB/s]
model.safetensors: 12%|███████▌ | 147M/1.19G [01:52<08:31, 2.04MB/s]
model.safetensors: 0%|▏ | 10.5M/3.71G [00:32<3:08:53, 326kB/s]
tokenizer.json: 43%|███████████████████████████▌ | 10.5M/24.3M [00:25<00:33, 419kB/s]
model.safetensors: 3%|█▉ | 115M/3.71G [01:49<44:21, 1.35MB/s]
model.safetensors: 0%|▎ | 21.0M/4.79G [00:36<2:11:27, 604kB/s]
model.safetensors: 4%|██▏ | 168M/4.79G [01:50<40:06, 1.92MB/s]
diffusion_pytorch_model.safetensors: 34%|███████████████ | 115M/337M [00:33<01:27, 2.53MB/s]
diffusion_pytorch_model.safetensors: 75%|████████████████████████████████▊ | 252M/337M [01:49<00:46, 1.85MB/s]
ACE-Step 1.5 Gradio UI 실행 로그 상세 문서
본 문서는 start_gradio_ui.bat 실행 시 출력되는 로그의 의미와 시스템 동작 흐름을 단계별로 설명한다.
환경 기준:
- OS: Windows
- Python: 3.10 (venv)
- GPU: RTX 4060 Laptop 8GB
- CUDA: torch 2.7.1 + cu128
1. 업데이트 확인 단계
[Update] Checking for updates...
[Update] Already up to date (34c52c2).
의미
- Git 기반 최신 버전 확인
34c52c2는 현재 체크아웃된 커밋 해시
정상 여부
정상. 업데이트 불필요.
2. Gradio 서버 초기화
Starting ACE-Step Gradio Web UI...
Server will be available at: http://127.0.0.1:7860
의미
- 로컬 웹 UI 실행 준비
- 7860 포트에서 서버 구동
접속 방법
브라우저에서 접속:
http://127.0.0.1:7860
3. 실행 환경 감지
[Environment] Embedded Python not found, checking for uv...
[Environment] Using uv package manager...
의미
- 내장 Python 미존재
- uv 패키지 매니저 사용
영향
문제 없음. 의존성 관리 도구 선택 단계.
4. 환경 설정 파일 로딩
Loaded configuration from .env.example (fallback)
의미
.env파일이 없어서.env.example사용
권장 조치
.env.example을 복사하여 .env 생성 후 사용자 설정 반영 권장
5. torchao 확장 모듈 경고
Skipping import of cpp extensions due to incompatible torch version
의미
- torch 2.7.1과 torchao 0.15.0 버전 불일치
- C++ 확장 비활성화
영향
- 일부 최적화 기능 비활성화
- 실행에는 문제 없음
6. bitsandbytes 경고
bitsandbytes not installed. Using standard AdamW.
의미
- 저비트 최적화 미사용
영향
- 학습 시 메모리 최적화 미적용
- 추론에는 큰 영향 없음
7. GPU 감지 및 자동 설정
CUDA GPU detected: RTX 4060 (8GB)
Configuration Tier: tier3
의미
- GPU 정상 인식
- 8GB 기준 자동 tier3 설정
자동 설정 내용
- LM 사용 시 최대 480초
- 배치 사이즈 2
- CPU Offload 자동 활성화
8. CPU Offload 활성화
Auto-enabling CPU offload (GPU 8GB < 20GB threshold)
의미
- VRAM 부족 방지 목적
- 일부 연산을 CPU로 이전
영향
- 속도 약간 감소
- OOM 방지
9. DiT 모델 초기화
Initializing DiT model: acestep-v15-turbo
DiT quantized with: int8_weight_only
DiT model initialized successfully
의미
- 메인 생성 모델 로딩 완료
- 8비트 양자화 적용
영향
- VRAM 절약
- 약간의 정밀도 감소 가능
10. 5Hz LM 초기화
Initializing 5Hz LM: acestep-5Hz-lm-0.6B
KV cache allocated: 1.91GB
Low GPU Memory Mode: True
5Hz LM initialized successfully
의미
- 6억 파라미터 LM 로딩
- GPU 메모리 40% 사용
- KV 캐시 약 1.9GB 할당
영향
- LM 기능 활성화
- 저메모리 모드 동작
11. 서비스 완료 및 서버 실행
Service initialization completed successfully!
Running on local URL: http://127.0.0.1:7860
의미
- 모든 모델 로딩 성공
- Gradio UI 정상 실행
실행 성공 판정 기준
다음 로그가 모두 확인되면 정상 실행으로 판단:
- CUDA GPU detected
- DiT model initialized successfully
- 5Hz LM initialized successfully
- Service initialization completed successfully
- Running on local URL
운영 시 참고 사항
VRAM 8GB 환경 특징
- CPU Offload 자동 활성화
- Low GPU Memory Mode 사용
- LM 사용 시 duration 제한 존재
메모리 부족 발생 시 조치
- LM 기능 비활성화
- duration 감소
- 배치 사이즈 감소
결론
본 로그는 RTX 4060 8GB 환경에서 ACE-Step 1.5 Gradio UI가 정상적으로 초기화 및 실행된 상태를 나타낸다.
경고 메시지는 존재하나 치명적 오류는 없으며, GPU 인식 및 모델 로딩이 모두 성공적으로 완료되었다.