HeartMuLa Studio をローカルで動かす (Suno 風 音楽生成AI 環境 / GeForce RTX 5060Ti 16GB)

cd \
git clone https://github.com/fspecii/HeartMuLa-Studio.git

cd HeartMuLa-Studio
py -3.11 -m venv venv
venv\Scripts\activate
python -m pip install --upgrade pip

pip install -U 'triton-windows>=3.2,<3.3'
pip install torchCodec
pip install -r backend/requirements.txt

pip uninstall -y torch torchvision torchaudio
pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu128

hf auth login

(venv) PS C:\HeartMuLa-Studio> hf auth login

    _|    _|  _|    _|    _|_|_|    _|_|_|  _|_|_|  _|      _|    _|_|_|      _|_|_|_|    _|_|      _|_|_|  _|_|_|_|
    _|    _|  _|    _|  _|        _|          _|    _|_|    _|  _|            _|        _|    _|  _|        _|
    _|_|_|_|  _|    _|  _|  _|_|  _|  _|_|    _|    _|  _|  _|  _|  _|_|      _|_|_|    _|_|_|_|  _|        _|_|_|
    _|    _|  _|    _|  _|    _|  _|    _|    _|    _|    _|_|  _|    _|      _|        _|    _|  _|        _|
    _|    _|    _|_|      _|_|_|    _|_|_|  _|_|_|  _|      _|    _|_|_|      _|        _|    _|    _|_|_|  _|_|_|_|

To log in, huggingface_hub requires a token generated from https://huggingface.co/settings/tokens .

Token can be pasted using ‘Right-Click’.
Enter your token (input will not be visible): トークンを入力
Add token as git credential? (Y/n) y
Token is valid (permission: read).
The token トークン名 has been saved to C:\Users\ユーザ名\.cache\huggingface\stored_tokens
Your token has been saved in your configured git credential helpers (helper-selector).
Your token has been saved to C:\Users\ユーザ名\.cache\huggingface\token
Login successful.
The current active token is: トークン名
(venv) PS C:\HeartMuLa-Studio>

HF_HEARTCODEC_REPO = "HeartMuLa/HeartCodec-oss"

HF_HEARTCODEC_REPO = "HeartMuLa/HeartCodec-oss-20260123"

    def _load_models(self):
        (
            mula_weights_path,
            mula_config_path,
            codec_path,
            tokenizer_path,
            gen_config_path,
        ) = _resolve_paths(
            self.ckpt_root,
            self.version,
            heartmula_weights_path=self.heartmula_weights_path,
        )

        from accelerate import init_empty_weights
        from mmgp import offload

        # Load tokenizer + gen config
        self.tokenizer = Tokenizer.from_file(str(tokenizer_path))
        self.gen_config = HeartMuLaGenConfig.from_file(str(gen_config_path))

        # Load HeartMuLa config
        with open(mula_config_path, encoding="utf-8") as fp:
            mula_config = HeartMuLaConfig(**json.load(fp))

        # Create empty model
        with init_empty_weights():
            self.mula = HeartMuLa(mula_config)

        # --- Add support for sharded HeartMuLa weights ---
        if mula_weights_path is None:
            sharded_mula = sorted(Path(self.ckpt_root).glob("model-*-of-*.safetensors"))
            if sharded_mula:
                mula_weights_path = [str(f) for f in sharded_mula]
            else:
                raise FileNotFoundError("HeartMuLa weights not found (no single file or shards detected)")

        # Quantized?
        is_quantized = "_int8" in str(mula_weights_path)
        if is_quantized:
            preprocess_fn = _dequantize_int8_state_dict
        else:
            preprocess_fn = _strip_heartmula_rope_cache

        # Load HeartMuLa weights
        offload.load_model_data(
            self.mula,
            mula_weights_path,
            default_dtype=None,
            writable_tensors=False,
            preprocess_sd=preprocess_fn,
        )

        # Decoder fix
        decoder = self.mula.decoder
        delattr(self.mula, "decoder")
        self.mula.decoder = [decoder]

        if hasattr(self.mula, "_interrupt_check"):
            self.mula._interrupt_check = self._abort_requested

        self.model = self.mula
        self.mula.eval()

        # Disable compilation
        self.mula.decoder[0].layers._compile_me = False
        self.mula.backbone.layers._compile_me = False

        first_param = next(self.mula.parameters(), None)
        if first_param is not None:
            self.mula_dtype = first_param.dtype

        # Resolve codec names
        codec_weights_name, codec_config_name = _resolve_codec_names(self.codec_version)

        # --- Add support for sharded HeartCodec weights ---
        if self.heartmula_weights_path and "quantized" in str(self.heartmula_weights_path):
            quantized_codec_path = Path(self.heartmula_weights_path).parent / "HeartMula_codec_int8.safetensors"
            if quantized_codec_path.is_file():
                codec_weights_path = quantized_codec_path
            else:
                quantized_codec_path = None
        else:
            quantized_codec_path = None

        if quantized_codec_path:
            codec_weights_path = quantized_codec_path
        else:
            sharded_codec = sorted(Path(codec_path).glob("model-*-of-*.safetensors"))
            if sharded_codec:
                codec_weights_path = [str(f) for f in sharded_codec]
            else:
                codec_weights_path = Path(codec_path) / codec_weights_name
                if not codec_weights_path.is_file():
                    codec_weights_path = Path(codec_path) / "model.safetensors"

        if not isinstance(codec_weights_path, list) and not codec_weights_path.is_file():
            raise FileNotFoundError(
                f"Expected HeartCodec weights at {codec_path}/{codec_weights_name} or model.safetensors but not found."
            )

        # Load codec config
        codec_config_path = Path(codec_path) / codec_config_name
        if not codec_config_path.is_file():
            codec_config_path = Path(codec_path) / "config.json"

        with open(codec_config_path, encoding="utf-8") as fp:
            codec_config = HeartCodecConfig(**json.load(fp))

        with init_empty_weights():
            self.codec = HeartCodec(codec_config)

        self.codec._offload_hooks = ["detokenize"]
        self.codec._model_dtype = self.VAE_dtype

        # Load codec weights
        offload.load_model_data(
            self.codec,
            codec_weights_path,
            default_dtype=self.VAE_dtype,
            writable_tensors=False,
            preprocess_sd=None,
        )

        self.codec.eval()

        first_param = next(self.codec.parameters(), None)
        if first_param is not None:
            self.codec_dtype = first_param.dtype

        self.sample_rate = getattr(self.codec, "sample_rate", 48000)
        self._offload_obj = None

pipeline.py を修正せずに手元で単一ファイルとして作り直す形でも動きました。
重みファイルを手元で結合するには以下で可能です。
venv を activate してから行う形です。

cd .\backend\models\HeartCodec-oss\
python -c "from safetensors.torch import load_file, save_file; import glob; w={}; [w.update(load_file(f)) for f in sorted(glob.glob('model-*.safetensors'))]; save_file(w,'HeartMula_codec.safetensors')"

cd C:\HeartMuLa-Studio\frontend
npm install
npm run build

(venv) PS C:\HeartMuLa-Studio> cd C:\HeartMuLa-Studio\frontend

(venv) PS C:\HeartMuLa-Studio\frontend> npm install

added 268 packages, and audited 269 packages in 6s

67 packages are looking for funding
run `npm fund` for details

12 vulnerabilities (1 moderate, 11 high)

To address issues that do not require attention, run:
npm audit fix

To address all issues (including breaking changes), run:
npm audit fix –force

Run `npm audit` for details.

(venv) PS C:\HeartMuLa-Studio\frontend> npm run build

> frontend@0.0.0 build
> tsc -b && vite build

vite v7.3.1 building client environment for production…
✓ 2159 modules transformed.
dist/index.html 0.49 kB │ gzip: 0.31 kB
dist/assets/index-nnSgdHv0.css 50.70 kB │ gzip: 8.38 kB
dist/assets/index-Bd-LrZYc.js 541.50 kB │ gzip: 163.49 kB

(!) Some chunks are larger than 500 kB after minification. Consider:
– Using dynamic import() to code-split the application
– Use build.rollupOptions.output.manualChunks to improve chunking: https://rollupjs.org/configuration-options/#output-manualchunks
– Adjust chunk size limit for this warning via build.chunkSizeWarningLimit.
✓ built in 2.92s

(venv) PS C:\HeartMuLa-Studio\frontend>

cd C:\HeartMuLa-Studio\
python -m uvicorn backend.app.main:app --host 0.0.0.0 --port 8000

(venv) PS C:\HeartMuLa-Studio> python -m uvicorn backend.app.main:app –host 0.0.0.0 –port 8000
W0223 14:45:43.557000 10876 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
INFO: Started server process [10876]
INFO: Waiting for application startup.
[Migration] Added generation_time_seconds column to job table
[Startup] downloading: 0% – Checking models…
[Startup] downloading: 5% – Preparing GPU…
[Startup] downloading: 10% – Downloading HeartMuLa-oss-3B-happy-new-year (~3GB)…
INFO: Application startup complete.
INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\file_download.py:986: UserWarning: `local_dir_use_symlinks` parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder as`local_dir`.
For more details, check out https://huggingface.co/docs/huggingface_hub/main/en/guides/download#download-files-to-local-folder.
warnings.warn(
config.json: 100%|████████████████████████████████████████████████████████████████████████████| 308/308 [00:00<?, ?B/s]
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
model.safetensors.index.json: 22.6kB [00:00, ?B/s]
README.md: 2.18kB [00:00, ?B/s].00B [00:00, ?B/s]
.gitattributes: 1.52kB [00:00, ?B/s]
model-00004-of-00004.safetensors: 100%|█████████████████████████████████████████████| 951M/951M [01:06<00:00, 14.4MB/s]
model-00002-of-00004.safetensors: 100%|███████████████████████████████████████████| 4.93G/4.93G [02:16<00:00, 36.0MB/s]
model-00001-of-00004.safetensors: 100%|███████████████████████████████████████████| 4.93G/4.93G [03:01<00:00, 27.2MB/s]
model-00003-of-00004.safetensors: 100%|███████████████████████████████████████████| 4.94G/4.94G [03:12<00:00, 25.7MB/s]
Fetching 8 files: 100%|██████████████████████████████████████████████████████████████████| 8/8 [03:13<00:00, 24.18s/it]
[Startup] downloading: 28% – HeartMuLa-oss-3B-happy-new-year downloaded███████████| 4.93G/4.93G [03:01<00:00, 44.5MB/s]
[Startup] downloading: 30% – Downloading HeartCodec (~1.5GB)…████████████▌ | 4.19G/4.94G [03:01<00:19, 38.5MB/s]
config.json: 100%|████████████████████████████████████████████████████████████████████████████| 981/981 [00:00<?, ?B/s]
model.safetensors.index.json: 83.9kB [00:00, ?B/s] | 0.00/981 [00:00<?, ?B/s]
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
Xet Storage is enabled for this repo, but the ‘hf_xet’ package is not installed. Falling back to regular HTTP download. For better performance, install the package with: `pip install huggingface_hub[hf_xet]` or `pip install hf_xet`
.gitattributes: 1.52kB [00:00, 540kB/s]
README.md: 100%|████████████████████████████████████████████████████████████████████████████| 31.0/31.0 [00:00<?, ?B/s]
model-00002-of-00002.safetensors: 100%|███████████████████████████████████████████| 1.71G/1.71G [00:38<00:00, 44.4MB/s]
model-00001-of-00002.safetensors: 100%|███████████████████████████████████████████| 4.93G/4.93G [01:49<00:00, 45.2MB/s]
Fetching 6 files: 100%|██████████████████████████████████████████████████████████████████| 6/6 [01:50<00:00, 18.36s/it]
[Startup] downloading: 38% – HeartCodec downloaded
[Startup] downloading: 39% – Downloading tokenizer…
tokenizer.json: 9.09MB [00:00, 219MB/s]
gen_config.json: 100%|████████████████████████████████████████████████████████████████████████| 101/101 [00:00<?, ?B/s]
[Startup] downloading: 40% – All models downloaded
[Startup] loading: 45% – Loading HeartMuLa model…

[Auto-Config] Detected 1 GPU(s):
GPU 0: NVIDIA GeForce RTX 5060 Ti (14.7GB free / 15.9GB total, SM 12.0) – ✓ Flash Attention
[Auto-Config] Using FREE VRAM (14.7GB) for configuration (total: 15.9GB)
[Auto-Config] Selected: MMGP BF16 mode (14.7GB VRAM)
[Auto-Config] Configuration: mmgp bf16 (Single GPU)

[Config] Using mmgp memory management (recommended for this GPU)
[GPU Config] Flash Attention ENABLED for NVIDIA GeForce RTX 5060 Ti (SM 12.0)
[mmgp] Using bf16 precision (faster)
[mmgp] Creating pipeline with balanced optimization…
[mmgp] Found sharded model weights (4 shards), using C:\HeartMuLa-Studio\backend\models\HeartMuLa-oss-3B-happy-new-year\model-00001-of-00004.safetensors
[mmgp] Setting up profile 3 for memory management…
************ Memory Management for the GPU Poor (mmgp 3.7.6) by DeepBeepMeep ************
[mmgp] Profiling enabled (profile 3)
[Startup] loading: 95% – Initializing pipeline…
[Startup] ready: 100% – Ready!

A tender and nostalgic reflection on the unwavering loyalty and quiet beauty of a Shiba Inu companion.

Japanese Folk, Ambient, Piano, Gentle, Melancholic

[Intro]
(Gentle, slightly melancholic piano chords begin, reminiscent of a Japanese folk song)
(Sound of a Shiba Inu panting softly)

[Verse 1]
雨上がりの空に　淡い光が降る
(Ameagari no sora ni　awai hikari ga furu)
散歩道に　足跡が残る
(Sanpo do ni　ashizuki ga nokoru)
忠実な瞳で　私を見つめる
(Chuujitsu na hitomi de　watashi o mitsumeru)
まるで天使　静かに微笑む
(Marude tenshi　shizuka ni hohoemu)

[Verse 2]
柴犬の毛並みは　太陽の金
(Shiba inu no kemuri wa　taiyou no kin)
賢い頭と　優しい心
(Kutsui no atama to　yasashii kokoro)
いつもそばに　寄り添う存在
(Itsumo soba ni　yorikonna zou sei)
小さな幸せを　教えてくれる
(Chiisana shiawase o　okoete kureru)

[Chorus]
柴犬のテーマソング　心に響く
(Shiba inu no temi soingu　kokoro ni hibiku)
温かい絆で　結ばれていく
(Atatakai kizuna de　musubareteiku)
忠誠の歌よ　永遠に
(Chuujitsu no uta yo　toen ni)
僕らの心に　刻まれよう
(Bokura no kokoro ni　kizamareyō)

[Bridge]
(Tempo slows slightly, a subtle string section enters)
まるでDrakeのように、人生は不確かな旅
(Marude Drake no you ni, jinsei wa fukakana tabi)
柴犬のように、今この瞬間を大切に
(Shiba inu no you ni, ima kono shunkan o taisetsu ni)
完璧じゃなくても、それが愛なのさ
(Kanpeki ja nakete mo, sore ga ai na nosa)

[Verse 3]
おやつをねだる　甘えた声
(Otsuke o negaru　ameta koe)
無邪気な笑顔で　私を癒す
(Muzukey na egao de　watashi o iyasu)
最高の相棒で　共に歩もう
(Saikou no aibō de　tomori ayou)
人生の岐路で　導いてくれる
(Jinsei no kiro de　michibiite kureru)

[Chorus]
柴犬のテーマソング　心に響く
(Shiba inu no temi soingu　kokoro ni hibiku)
温かい絆で　結ばれていく
(Atatakai kizuna de　musubareteiku)
忠誠の歌よ　永遠に
(Chuujitsu no uta yo　toen ni)
僕らの心に　刻まれよう
(Bokura no kokoro ni　kizamareyō)

[Outro]
(Piano chords return, fading slowly)
(Sound of a Shiba Inu whimpering softly)
柴犬…　ずっと…
(Shiba inu… zutto…)
(Final piano chord rings out)

Failed to load Heartlib model: 401 Client Error. (Request ID: Root=xxxx)

Repository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication
Invalid username or password.
[Startup] error: 0% – Failed to load model: 401 Client Error. (Request ID: Root=xxxx)

Repository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication
Invalid username or password.
Task exception was never retrieved
future: exception=RepositoryNotFoundError(‘401 Client Error. (Request ID: Root=xxxx)\n\nRepository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.\nPlease make sure you specified the correct repo_id and repo_type.\nIf you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication\nInvalid username or password.’)>
Traceback (most recent call last):
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_http.py”, line 403, in hf_raise_for_status
response.raise_for_status()
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\requests\models.py”, line 1026, in raise_for_status
raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1212, in initialize_with_progress
raise e
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1181, in initialize_with_progress
model_path = await self._download_models_with_progress(DEFAULT_MODEL_DIR, version)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1122, in _download_models_with_progress
await loop.run_in_executor(
File “C:\Program Files\Python311\Lib\concurrent\futures\thread.py”, line 58, in run
result = self.fn(*self.args, *self.kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1124, in lambda: snapshot_download( ^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, *kwargs) ^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub_snapshot_download.py”, line 245, in snapshot_download raise api_call_error File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub_snapshot_download.py”, line 165, in snapshot_download repo_info = api.repo_info(repo_id=repo_id, repo_type=repo_type, revision=revision) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, *kwargs) ^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\hf_api.py”, line 2867, in repo_info return method( ^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\hf_api.py”, line 2661, in model_info
hf_raise_for_status(r)
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_http.py”, line 453, in hf_raise_for_status
raise _format(RepositoryNotFoundError, message, response) from e
huggingface_hub.errors.RepositoryNotFoundError: 401 Client Error. (Request ID: Root=xxxx)

Repository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication
Invalid username or password.

[Startup] error: 0% – Failed to load model: 404 Client Error. (Request ID: Root=xxxx)

Repository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication
Task exception was never retrieved
future: exception=RepositoryNotFoundError(‘404 Client Error. (Request ID: Root=xxxx)\n\nRepository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.\nPlease make sure you specified the correct repo_id and repo_type.\nIf you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication’)>
Traceback (most recent call last):
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_http.py”, line 403, in hf_raise_for_status
response.raise_for_status()
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\requests\models.py”, line 1026, in raise_for_status
raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 404 Client Error: Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1212, in initialize_with_progress
raise e
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1181, in initialize_with_progress
model_path = await self._download_models_with_progress(DEFAULT_MODEL_DIR, version)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1122, in _download_models_with_progress
await loop.run_in_executor(
File “C:\Program Files\Python311\Lib\concurrent\futures\thread.py”, line 58, in run
result = self.fn(*self.args, *self.kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1124, in lambda: snapshot_download( ^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, *kwargs) ^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub_snapshot_download.py”, line 245, in snapshot_download raise api_call_error File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub_snapshot_download.py”, line 165, in snapshot_download repo_info = api.repo_info(repo_id=repo_id, repo_type=repo_type, revision=revision) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, *kwargs) ^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\hf_api.py”, line 2867, in repo_info return method( ^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_validators.py”, line 114, in _inner_fn return fn(args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\hf_api.py”, line 2661, in model_info
hf_raise_for_status(r)
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\huggingface_hub\utils_http.py”, line 453, in hf_raise_for_status
raise _format(RepositoryNotFoundError, message, response) from e
huggingface_hub.errors.RepositoryNotFoundError: 404 Client Error. (Request ID: Root=xxxx)

Repository Not Found for url: https://huggingface.co/api/models/HeartMuLa/HeartCodec-oss/revision/main.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https://huggingface.co/docs/huggingface_hub/authentication

…

[Quantization] Loading HeartMuLa with 4-bit NF4 quantization…
torch_dtype is deprecated! Use dtype instead!
Failed to load Heartlib model: Torch not compiled with CUDA enabled
[Startup] error: 0% – Failed to load model: Torch not compiled with CUDA enabled
Task exception was never retrieved
future: exception=AssertionError(‘Torch not compiled with CUDA enabled’)>
Traceback (most recent call last):
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1212, in initialize_with_progress
raise e
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1196, in initialize_with_progress
self.pipeline = await loop.run_in_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\Program Files\Python311\Lib\concurrent\futures\thread.py”, line 58, in run
result = self.fn(*self.args, *self.kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1198, in lambda mp=model_path, v=version: self.load_pipeline_multi_gpu(mp, v) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1430, in _load_pipeline_multi_gpu pipeline = create_quantized_pipeline( ^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 690, in create_quantized_pipeline heartmula = HeartMuLa.from_pretrained( ^^^^^^^^^^^^^^^^^^^^^^^^^^ File “C:\HeartMuLa-Studio\venv\Lib\site-packages\transformers\modeling_utils.py”, line 277, in _wrapper return func(args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\transformers\modeling_utils.py”, line 5051, in from_pretrained
) = cls._load_pretrained_model(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\transformers\modeling_utils.py”, line 5435, in _load_pretrained_model
caching_allocator_warmup(model, expanded_device_map, hf_quantizer)
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\transformers\modeling_utils.py”, line 6092, in caching_allocator_warmup
index = device.index if device.index is not None else torch_accelerator_module.current_device()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\venv\Lib\site-packages\torch\cuda__init_.py”, line 878, in current_device
lazy_init() File “C:\HeartMuLa-Studio\venv\Lib\site-packages\torch\cuda__init_.py”, line 305, in _lazy_init
raise AssertionError(“Torch not compiled with CUDA enabled”)
AssertionError: Torch not compiled with CUDA enabled

[mmgp] Found sharded model weights (4 shards), using C:\HeartMuLa-Studio\backend\models\HeartMuLa-oss-3B-happy-new-year\model-00001-of-00004.safetensors
Failed to load Heartlib model: Expected HeartCodec weights at C:\HeartMuLa-Studio\backend\models\HeartCodec-oss/HeartMula_codec.safetensors or model.safetensors but not found.
[Startup] error: 0% – Failed to load model: Expected HeartCodec weights at C:\HeartMuLa-Studio\backend\models\HeartCodec-oss/HeartMula_codec.safetensors or model.safetensors but not found.
Task exception was never retrieved
future: exception=FileNotFoundError(‘Expected HeartCodec weights at C:\\HeartMuLa-Studio\backend\models\HeartCodec-oss/HeartMula_codec.safetensors or model.safetensors but not found.’)>
Traceback (most recent call last):
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1212, in initialize_with_progress
raise e
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1196, in initialize_with_progress
self.pipeline = await loop.run_in_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\Program Files\Python311\Lib\concurrent\futures\thread.py”, line 58, in run
result = self.fn(*self.args, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1198, in
lambda mp=model_path, v=version: self._load_pipeline_multi_gpu(mp, v)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 1372, in _load_pipeline_multi_gpu
pipeline = create_mmgp_pipeline(
^^^^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\app\services\music_service.py”, line 601, in create_mmgp_pipeline
pipeline = HeartMuLaPipeline(
^^^^^^^^^^^^^^^^^^
File “C:\HeartMuLa-Studio\backend\heartmula\pipeline.py”, line 295, in init
self._load_models()
File “C:\HeartMuLa-Studio\backend\heartmula\pipeline.py”, line 374, in _load_models
raise FileNotFoundError(
FileNotFoundError: Expected HeartCodec weights at C:\HeartMuLa-Studio\backend\models\HeartCodec-oss/HeartMula_codec.safetensors or model.safetensors but not found.