Release WildSongBench

.gitattributes

@@ -0,0 +1,60 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,141 @@
+---
+pretty_name: WildSongBench
+language:
+- zh
+- en
+size_categories:
+- n<1K
+tags:
+- music-generation
+- benchmark
+configs:
+- config_name: default
+  data_files:
+  - split: test
+    path: prompts.jsonl
+---
+
+<h1 align="center">🤗 WildSongBench</h1>
+<p align="center"><strong>A benchmark for full-song music generation</strong></p>
+<p align="center">192 prompts · 94 Chinese · 98 English</p>
+
+<p align="center">
+  <a href="https://map-yue2.github.io/">🎵&nbsp;YuE2&nbsp;project</a>
+  ·
+  <a href="#quick-start">🚀&nbsp;Quick&nbsp;start</a>
+  ·
+  <a href="#benchmarks">📊&nbsp;Benchmarks</a>
+  ·
+  <a href="#reproduce-standard-yue2">🔁&nbsp;Reproduce</a>
+  ·
+  <a href="#citation">📚&nbsp;Citation</a>
+</p>
+<p align="center">
+  <a href="https://huggingface.co/m-a-p/YuE2-3B"><img alt="🤗 YuE2-3B" src="https://img.shields.io/badge/YuE2--3B-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/m-a-p/YuE2-Vae"><img alt="🤗 YuE2-Vae" src="https://img.shields.io/badge/YuE2--Vae-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/m-a-p/YuE2-Vae-legacy"><img alt="🤗 YuE2-Vae-legacy" src="https://img.shields.io/badge/YuE2--Vae--legacy-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/m-a-p/MERT-v2-30s"><img alt="🤗 MERT-v2-30s" src="https://img.shields.io/badge/MERT--v2--30s-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/m-a-p/MERT-v2-FullSong"><img alt="🤗 MERT-v2-FullSong" src="https://img.shields.io/badge/MERT--v2--FullSong-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/datasets/m-a-p/WildSongBench"><img alt="🤗 WildSongBench" src="https://img.shields.io/badge/WildSongBench-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+  &nbsp;
+  <a href="https://huggingface.co/m-a-p/SheetSage2"><img alt="SheetSage2" src="https://img.shields.io/badge/SheetSage2-374151?logo=huggingface&amp;logoColor=FFD21E" height="20" /></a>
+</p>
+
+WildSongBench (WSB) contains **192 song-generation prompts: 94 Chinese and 98 English**, used in the [YuE2 benchmarks](https://huggingface.co/m-a-p/YuE2-3B#benchmarks). This repository provides prompts, exact inputs and seeds, reference scores, and a minimal evaluator for **SongBench, Q3O, MuQ-MuLan, AllMusicCaps and Qwen3-ASR PER**. Audio is not included.
+
+<a id="dataset"></a>
+<a id="quick-start"></a>
+
+## 🚀 Quick start
+
+```python
+from datasets import load_dataset
+
+wsb = load_dataset("m-a-p/WildSongBench", split="test")
+```
+
+Each record contains a `prompt`. [Generation inputs](benchmark/reproduction_manifest.jsonl) contain the corresponding style, lyrics and exact seeds for every prompt.
+
+<a id="benchmark-results"></a>
+<a id="benchmarks"></a>
+
+## 📊 Benchmarks
+
+Recorded paper results on all 192 prompts. SongBench averages seven quality dimensions; Q3O measures prompt control on a 0–5 scale. MuQ-MuLan and AllMusicCaps measure audio–text similarity. PER is phoneme error rate.
+
+| Model | Access | SongBench ↑ | Q3O ↑ | MuQ-MuLan ↑ | AllMusicCaps ↑ | PER ↓ |
+|---|---|---:|---:|---:|---:|---:|
+| YuE2 | Open weights | 6.73 | 4.68 | 0.507 | 0.405 | 8.44% |
+| YuE2 (best-of-8) | Open weights | 6.96 | 4.70 | 0.505 | 0.398 | 9.79% |
+| Mureka 9 | Proprietary | 6.94 | 4.64 | 0.439 | 0.410 | 11.69% |
+| Suno v5 | Proprietary | 6.87 | 4.59 | 0.543 | 0.435 | 8.10% |
+| Suno v5.5 | Proprietary | 6.71 | 4.59 | 0.509 | 0.392 | 5.96% |
+| Suno v4.5 | Proprietary | 6.70 | 4.41 | 0.502 | 0.387 | 5.80% |
+| LeVo 2 | Open weights | 6.32 | 3.95 | 0.354 | 0.268 | 26.12% |
+| MiniMax Music 2.6 | Proprietary | 6.32 | 4.57 | 0.425 | 0.367 | 24.55% |
+| MiniMax Music 3 | Open weights | 6.28 | 4.44 | 0.393 | 0.361 | 6.27% |
+| HeartMuLa | Open weights | 6.25 | 3.49 | 0.382 | 0.279 | 10.71% |
+| Muse | Open weights | 6.03 | 4.40 | 0.394 | 0.347 | 33.42% |
+| ACE-Step 1.5 | Open weights | 6.01 | 4.58 | 0.437 | 0.387 | 7.46% |
+| DiffRhythm 2 | Open weights | 5.24 | 4.09 | 0.378 | 0.325 | 18.41% |
+| YuE 1 | Open weights | 4.92 | 3.73 | 0.262 | 0.288 | 36.38% |
+| SongBloom | Open weights | 4.24 | 3.03 | 0.270 | 0.193 | 19.19% |
+
+Systems use their recorded candidate-selection protocols; this is not a matched-compute comparison. [Full results and protocols](benchmark/comparison_summary.csv).
+
+<a id="reproduce-standard-yue2"></a>
+
+## 🔁 Reproduce standard YuE2
+
+Use [YuE2-3B](https://huggingface.co/m-a-p/YuE2-3B) with **[YuE2-Vae-legacy](https://huggingface.co/m-a-p/YuE2-Vae-legacy)** for benchmark scoring. The scripts pin the tested checkpoints and inference version. The benchmark recipe retains inference version 0.1.3 with updated public repository defaults; model weights and numerical inference code are unchanged.
+
+Run from the downloaded repository in a fresh Python 3.12 environment on Linux, with a C/C++ compiler for Q3O's key scorer. **Q3O requires an 80 GB GPU; the other metrics support 24 GB.** Public evaluator weights download once at fixed revisions (~83 GB for all metrics). The installer creates the required environments. Select your GPU by its numeric index below.
+
+```bash
+python -m pip install huggingface-hub==0.36.2
+export CUDA_VISIBLE_DEVICES=0
+python code/install_inference.py
+tar --no-same-owner -xzf evaluation/wsb_metrics.tar.gz
+python wsb_metrics/wsb_eval.py prepare \
+  --metrics all --download huggingface --install
+
+python code/generate.py --candidates 2 --backend torch --output runs/wsb
+python wsb_metrics/wsb_eval.py score \
+  --input runs/wsb/evaluation_inputs.jsonl \
+  --output runs/scores --mode standard --metrics per,songbench
+python wsb_metrics/wsb_eval.py score \
+  --input runs/wsb/evaluation_inputs.jsonl \
+  --per-results runs/scores/per_song.jsonl \
+  --output runs/additional --metrics mulan,allmusiccaps,q3o
+```
+
+For prompt row `i` (zero-based, 0–191), generate two candidates with seeds `831001 + i` and `831019 + i`. The script applies each seed to symbolic planning, semantic generation and acoustic synthesis. Four ASR passes estimate each candidate's PER; the lower-PER candidate is selected for every metric. Score all 192 prompts and retain capped outputs. Q3O uses the paper's fixed candidate weights and traditional key scoring. The scripts handle preprocessing and tie breaks.
+
+On a 24 GB GPU, prepare `per,songbench,mulan,allmusiccaps` and omit `q3o` from the last command. The second scoring command reuses the first run's PER selection and evaluates exactly those 192 songs.
+
+Selected scores are in `runs/scores/summary.json` (SongBench/PER) and `runs/additional/summary.json` (other metrics), under `standard_selected.groups.all`.
+
+Fresh SongBench scores are **6.73 (H800)** and **6.74 (RTX 4090)**, against **6.73** in the paper. On the original 192 songs, MuQ-MuLan and AllMusicCaps mean scores match within 0.000001. Q3O gives **4.69 vs 4.68** on the original audio and **4.71** on fresh H800 generations; equality to two decimals remains unresolved. Best-of-8 has not been rerun in full.
+
+<a id="citation"></a>
+
+## 📚 Citation
+
+**Technical report coming soon.** For now, please cite [YuE](https://arxiv.org/abs/2503.08638) when using WildSongBench in your research.
+
+```bibtex
+@article{yuan2025yue,
+  title = {{YuE}: Scaling Open Foundation Models for Long-Form Music Generation},
+  author = {Yuan, Ruibin and Lin, Hanfeng and Guo, Shuyue and Zhang, Ge and Pan, Jiahao and Zang, Yongyi and Liu, Haohe and Liang, Yiming and Ma, Wenye and Du, Xingjian and Du, Xinrun and Ye, Zhen and Zheng, Tianyu and Jiang, Zhengxuan and Ma, Yinghao and Liu, Minghao and Tian, Zeyue and Zhou, Ziya and Xue, Liumeng and Qu, Xingwei and Li, Yizhi and Wu, Shangda and Shen, Tianhao and Ma, Ziyang and Zhan, Jun and Wang, Chunhui and Wang, Yatian and Chi, Xiaowei and Zhang, Xinyue and Yang, Zhenzhu and Wang, Xiangzhou and Liu, Shansong and Mei, Lingrui and Li, Peng and Wang, Junjie and Yu, Jianwei and Pang, Guojian and Li, Xu and Wang, Zihao and Zhou, Xiaohuan and Yu, Lijun and Benetos, Emmanouil and Chen, Yong and Lin, Chenghua and Chen, Xie and Xia, Gus and Zhang, Zhaoxiang and Zhang, Chao and Chen, Wenhu and Zhou, Xinyu and Qiu, Xipeng and Dannenberg, Roger and Liu, Jiaheng and Yang, Jian and Huang, Wenhao and Xue, Wei and Tan, Xu and Guo, Yike},
+  journal = {arXiv preprint arXiv:2503.08638},
+  year = {2025},
+  eprint = {2503.08638},
+  archivePrefix = {arXiv},
+  url = {https://arxiv.org/abs/2503.08638}
+}
+```

SHA256SUMS

@@ -0,0 +1,7 @@
+05ab7a768d2f01b9f4d87882357eec68da1b535a017d379d55187745cbbb2d96  README.md
+260a66760ddfcf10b3db48f150d0f905dbc777ee33af27789b69c98b160ac70c  benchmark/comparison_summary.csv
+859e5225d319912efb64d587d1d5eb6cf0abc9fb162cc3d5f02d1688cff2edc0  benchmark/reproduction_manifest.jsonl
+41e069ed0f9d7fde27b277de40fe69d31fd0f5eed7a481b0af00e6486a6229c1  code/generate.py
+872a7c3e4a3ac5f842847f2909b82de59221d399ef3879a8310249f3ae11d81a  code/install_inference.py
+bd550b36f1fc43cad3254a9897dd3a078e50c4eb9e8d4f295ac4b5e81d1fc4fd  evaluation/wsb_metrics.tar.gz
+1b554a824eba8e02eabe02299e48ed2215764931b8f05535cf8c6d250ddf8326  prompts.jsonl

benchmark/comparison_summary.csv

@@ -0,0 +1,16 @@
+system,n,protocol,per,control_overall,songbench_global_avg,songeval_global_avg,audiobox_pq,mulan,allmusiccaps,audiobox_ce,audiobox_cu,audiobox_pc
+YuE2,192,roll2 / four-ASR lower-PER selection,0.08443630690279864,4.681896986601725,6.731599404761909,4.262452916666666,8.25979095697403,0.5068482221880307,0.40536119728009606,7.5946645612518,7.883815668523312,6.329105652868748
+YuE2 (best-of-8),192,best-of-8 / Musicality → Control → PER,0.09786539961704843,4.70087253751977,6.963229315476184,4.296048333333334,8.271365446348986,0.5050805467180908,0.398045457637636,7.585739687085152,7.882911428809166,6.349051711459954
+ACE-Step 1.5,192,roll2 / four-ASR lower-PER selection,0.07457884998484957,4.580894501177339,6.011793824404761,3.846484166666666,8.051833018660545,0.43723272854307044,0.38685302749703016,7.345652607580026,7.696453630924225,6.559373706579208
+DiffRhythm 2,192,roll2 / four-ASR lower-PER selection,0.18413265339015286,4.087044602340498,5.242845982142859,3.5244677083333333,7.978225273390611,0.37817035724098486,0.32549789230688475,7.221992149949074,7.5754738847414655,6.354656022042036
+HeartMuLa,192,roll2 / four-ASR lower-PER selection,0.10705881594471056,3.490655976705673,6.248347544642855,4.551859166666667,8.293305036922296,0.3823466453177389,0.2786190857962841,7.695213384926319,7.913336729009946,6.434617171684901
+LeVo 2,192,roll2 / four-ASR lower-PER selection,0.26116136186343974,3.9457878565807953,6.324671800595234,4.023369479166667,8.396623315910498,0.3542381529114209,0.2679880544101252,7.640917373200257,7.93210681527853,6.627885211259127
+MiniMax Music 2.6,192,delivered candidate / stored ASR selection,0.24545647955335212,4.568809841873251,6.3221788690476215,4.098453333333334,8.171058195332686,0.42510899043797207,0.3669951342102043,7.385788751145204,7.8092692494392395,6.509756391247113
+MiniMax Music 3,192,roll2 / four-ASR lower-PER selection / Standard + Official Rewriter,0.06268414232388418,4.43616720498247,6.282988392857145,4.099352187500001,8.28245102117459,0.39279851930526394,0.36085424468425725,7.516246639192104,7.886647321283817,6.417244121432304
+Mureka 9,192,delivered candidate / stored ASR selection,0.11690337887061523,4.636807564214055,6.937748809523812,4.411131979166667,8.022609589000544,0.43942587031051517,0.41017753163275,7.425450682640076,7.709370714922746,6.665880647798379
+Muse,192,roll2 / four-ASR lower-PER selection,0.33418648580693827,4.403758449771435,6.034903943452382,3.788667708333335,8.051745663086573,0.3936510290513979,0.3465821466803997,7.357404105365276,7.562788394590219,6.79859949896733
+SongBloom,192,roll2 / four-ASR lower-PER selection; fixed prompt audio,0.19193351857701904,3.0286910546158463,4.235043750000001,3.2051130208333345,8.153916046023369,0.2697471845106823,0.19264957021005102,7.1598992844422655,7.663380955656369,4.881139911711216
+Suno v4.5,192,delivered candidate / stored ASR selection,0.05798428222369787,4.41490242329515,6.699528199404766,4.366563958333332,8.254062737027803,0.5021767822715143,0.38734816436772235,7.566704971094926,7.847558493415515,6.4596482912699384
+Suno v5,192,delivered candidate / stored ASR selection,0.080993611026942,4.5907186541713685,6.872074330357141,4.357936874999999,8.169838989774385,0.5428044088184834,0.43525151138116297,7.522664589186509,7.8013999834656715,6.551953633626302
+Suno v5.5,192,delivered candidate / stored ASR selection,0.05958313723720433,4.591446755445953,6.714983407738093,4.2151607291666675,8.195489337046942,0.5088621161606474,0.3917035845806822,7.4402403980493546,7.9158454810579615,6.531025744974613
+YuE 1,192,roll2 / four-ASR lower-PER selection,0.3637846445448743,3.7300764989993715,4.916463244047622,3.214967395833334,7.868339642882347,0.262323199029197,0.2881557388876293,6.952277842909098,7.441528876622518,5.72498715420564

code/generate.py

@@ -0,0 +1,109 @@
+#!/usr/bin/env python3
+"""Generate WSB candidates with pinned Hugging Face releases; never reuse audio."""
+import argparse
+import hashlib
+import importlib.metadata
+import json
+import platform
+from pathlib import Path
+
+MODEL = "m-a-p/YuE2-3B"
+MODEL_REVISION = "1a96eca688d6ae5d7f0feb88573fec89920fcd19"
+VAE = "m-a-p/YuE2-Vae-legacy"
+VAE_REVISION = "b54118f0fc462f08999d1ec07e88817f4ee3f770"
+MODEL_SHA = "1d55c42c1a9875c34f5d736e15078449992b044e807ce2a138e6cf289a1e59e9"
+VAE_SHA = "b6d283628913bb41145ba99e2314eef613905ee95f690eb70e8212d5f4965044"
+
+
+def sha(path):
+    h = hashlib.sha256()
+    with Path(path).open("rb") as stream:
+        for block in iter(lambda: stream.read(8 << 20), b""):
+            h.update(block)
+    return h.hexdigest()
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--manifest", type=Path, default=Path(__file__).resolve().parents[1] / "benchmark/reproduction_manifest.jsonl")
+    parser.add_argument("--output", type=Path, required=True)
+    parser.add_argument("--candidates", type=int, choices=(2, 8), default=8)
+    parser.add_argument("--backend", choices=("torch", "torch-eager", "vllm"), default="torch")
+    parser.add_argument("--shard-index", type=int, default=0)
+    parser.add_argument("--num-shards", type=int, default=1)
+    parser.add_argument("--prompt-index", type=int, action="append", help="Optional diagnostic subset; repeat for multiple row indices")
+    parser.add_argument("--cache-dir", type=Path)
+    args = parser.parse_args()
+    if not 0 <= args.shard_index < args.num_shards:
+        parser.error("Require 0 <= shard-index < num-shards")
+    if args.prompt_index is not None and (len(set(args.prompt_index)) != len(args.prompt_index)
+            or any(not 0 <= i < 192 for i in args.prompt_index)):
+        parser.error("Prompt indices must be distinct integers from 0 to 191")
+    if args.output.exists() and any(args.output.iterdir()):
+        parser.error("Use a new output directory so previous candidates cannot be reused or overwritten")
+    all_rows = [json.loads(line) for line in args.manifest.read_text(encoding="utf-8").splitlines() if line.strip()]
+    assert len(all_rows) == 192 and [r["prompt_index"] for r in all_rows] == list(range(192))
+    bases = [831001, 831019, 831037, 831061, 831083, 831109, 831127, 831149]
+    for row in all_rows:
+        assert row["candidate_ar_seeds"] == [s + row["prompt_index"] for s in bases]
+        assert row["candidate_ar_seeds"] == row["candidate_nar_seeds"]
+    rows = [r for r in all_rows if r["prompt_index"] % args.num_shards == args.shard_index]
+    if args.prompt_index is not None:
+        rows = [r for r in rows if r["prompt_index"] in args.prompt_index]
+    assert rows
+    import torch
+    from yue2 import YuE2Pipeline
+    from yue2.protocol import SongRequest
+    from yue2.storage import verify_result
+    assert importlib.metadata.version("yue2-infer") == "0.1.3"
+    args.output.mkdir(parents=True, exist_ok=True)
+    evaluation_inputs = []
+    for row in rows:
+        for candidate_index, seed in enumerate(row["candidate_ar_seeds"][:args.candidates]):
+            ident = f"wsb_{row['prompt_index']:03d}_r{candidate_index}_s{seed}"
+            evaluation_inputs.append({"prompt_index": row["prompt_index"],
+                                      "candidate_index": candidate_index, "seed": seed,
+                                      "path": f"songs/{ident}"})
+    (args.output / "evaluation_inputs.jsonl").write_text(
+        "".join(json.dumps(row) + "\n" for row in evaluation_inputs))
+    report = {"model": MODEL, "model_revision": MODEL_REVISION, "vae": VAE,
+              "vae_revision": VAE_REVISION, "backend": args.backend,
+              "manifest_sha256": sha(args.manifest), "python": platform.python_version(),
+              "torch": torch.__version__, "gpu": torch.cuda.get_device_name(0),
+              "shard_index": args.shard_index, "num_shards": args.num_shards,
+              "prompt_indices": [r["prompt_index"] for r in rows],
+              "expected_candidates": len(rows) * args.candidates, "results": [], "complete": False}
+    with YuE2Pipeline.from_pretrained(MODEL, revision=MODEL_REVISION, vae=VAE,
+            vae_revision=VAE_REVISION, cache_dir=args.cache_dir,
+            backend=args.backend, device="cuda:0", memory_budget_gib=24) as pipe:
+        assert sha(pipe.model_dir / "model.safetensors") == MODEL_SHA
+        assert sha(pipe.vae_dir / "model.safetensors") == VAE_SHA
+        report["generation_config"] = pipe.generation_config.to_dict()
+        for row in rows:
+            for candidate_index, seed in enumerate(row["candidate_ar_seeds"][:args.candidates]):
+                ident = f"wsb_{row['prompt_index']:03d}_r{candidate_index}_s{seed}"
+                request = dict(id=ident, style=row["style"], lyrics=row["lyrics"],
+                               cot="full", cfg_scale=1.0, seed=seed)
+                assert SongRequest(**request).text() == row["generation_prompt"]
+                result = {"id": ident, "prompt_index": row["prompt_index"],
+                          "candidate_index": candidate_index, "seed": seed}
+                try:
+                    song = pipe(**request)
+                    song.save_artifacts(args.output / "songs" / ident)
+                    receipt = verify_result(args.output / "songs" / ident)
+                    result.update(status="complete", identity=receipt["identity"],
+                                  truncated=receipt["truncated"], audio_seconds=receipt["audio_seconds"])
+                    del song
+                except Exception as exc:
+                    result.update(status="failed", error=f"{type(exc).__name__}: {exc}")
+                report["results"].append(result)
+                (args.output / "generation_report.json").write_text(json.dumps(report, indent=2) + "\n")
+                print(json.dumps(result), flush=True)
+    report["complete"] = (len(report["results"]) == report["expected_candidates"]
+                          and all(r["status"] == "complete" for r in report["results"]))
+    (args.output / "generation_report.json").write_text(json.dumps(report, indent=2) + "\n")
+    return 0 if report["complete"] else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

code/install_inference.py

@@ -0,0 +1,19 @@
+#!/usr/bin/env python3
+"""Install the exact inference wheel used by this reproduction recipe."""
+import hashlib
+import argparse
+import subprocess
+import sys
+from huggingface_hub import hf_hub_download
+
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument("--fast", action="store_true", help="Also install the released optional vLLM dependencies")
+args = parser.parse_args()
+wheel = hf_hub_download(
+    "m-a-p/YuE2-3B", "yue2_infer-0.1.3-py3-none-any.whl",
+    revision="1a96eca688d6ae5d7f0feb88573fec89920fcd19",
+)
+with open(wheel, "rb") as stream:
+    assert hashlib.sha256(stream.read()).hexdigest() == "156d6b8b7e46c3bd23cf22a50d3c20c403e5dc7f4a315582a224a840b01f4f07"
+subprocess.check_call([sys.executable, "-m", "pip", "install", "--force-reinstall", "--no-deps", wheel])
+subprocess.check_call([sys.executable, "-m", "pip", "install", wheel + ("[fast]" if args.fast else "")])

evaluation/wsb_metrics.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd550b36f1fc43cad3254a9897dd3a078e50c4eb9e8d4f295ac4b5e81d1fc4fd
+size 505834965