evaluation
Evaluate Spleeter models on MUSDB-style datasets, dependency gates, output metrics, fixtures, and evaluation troubleshooting.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- evaluation
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Evaluate Spleeter models on MUSDB-style datasets, dependency gates, output metrics, fixtures, and evaluation troubleshooting.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/evaluation-workflow.md6.0 KB
- 参考文档references/musdb-format-and-metrics.md4.5 KB
- 参考文档references/troubleshooting.md5.7 KB
- 脚本scripts/create_eval_fixture.py6.5 KB
03 SKILL.md 正文
Spleeter evaluation router
Use this sub-skill when the task mentions spleeter evaluate, MUSDB or MUSDB-like test sets, musdb/museval, SDR/SAR/SIR/ISR, evaluation output directories, fake evaluation fixtures, or evaluation failures related to model downloads and caches.
Do not use this sub-skill for generic source separation recipes; route those to separation. Do not use it for training CSV/config construction except for a tiny evaluation fixture; route training work to training.
Fast route
| User intent | Go to |
|---|---|
| Install or check evaluation prerequisites | Evaluation workflow and root installation/runtime |
Run spleeter evaluate on a MUSDB-style dataset |
Evaluation workflow |
Understand mus_dir layout or distinguish ground truth from estimates |
MUSDB format and metrics |
| Interpret SDR/SAR/SIR/ISR metrics or compiled medians | Metric compilation behavior |
| Make a tiny deterministic fixture for smoke/layout checks | scripts/create_eval_fixture.py and fixture notes |
| Debug exit code 10, missing stems, cache/network, JSON metrics, or slow runs | Evaluation troubleshooting |
| Compare evaluation CLI flags with the global CLI catalog | Root CLI reference |
| Select model descriptors, local configs, or cache/network policy | Root models/configuration |
Core command shape
spleeter evaluate \
--mus_dir MUSDB_ROOT \
--output_path OUTPUT \
--params_filename spleeter:4stems \
--adapter spleeter.audio.ffmpeg.FFMPEGProcessAudioAdapter \
--mwf \
--verbose
python -m spleeter evaluate ... is equivalent when the console entry point is unavailable. Short forms supported by the CLI include -o/--output_path, -p/--params_filename, and -a/--adapter. --mus_dir, --mwf, and --verbose are long-form options.
Keep these boundaries clear
MUSDB_ROOT/test/<song>/mixture.wavis the mixture input that Spleeter separates.MUSDB_ROOT/test/<song>/{vocals,drums,bass,other}.wavare ground-truth source files required bymuseval; they are not Spleeter outputs.- Evaluation writes Spleeter estimates under
OUTPUT/audio/test/<song>/and metric JSON files underOUTPUT/metrics/test/. - Base Spleeter installs may not include evaluation dependencies. If
musdbormusevalcannot import, Spleeter logs the missing extras and exits with status10. - GPU use is optional acceleration only; the verified runtime path is TensorFlow CPU.
Evidence basis: spleeter/__main__.py, spleeter/options.py, tests/test_eval.py, pyproject.toml, README.md, and configs/musdb_config.json.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).