{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "chopin-in-five-minutes",
 "title": "Teaching a Music Model Chopin in Five Minutes",
 "dek": "Nineteen public-domain Chopin recordings, four minutes and forty-nine seconds of training on a five-year-old graphics card, and a dial you can hear — five clips on the page that differ from each other in exactly one way.",
 "published": "2026-09-01",
 "series": [
  "guide",
  "bench"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/chopin-in-five-minutes/",
  "md_url": "https://research.strata2signal.com/chopin-in-five-minutes/index.md",
  "md_sha": "f92c6f5a709c350dd87b38f1013661226283758d9ef09fc1f2a552875c97cf1f",
  "html_sha": "6824413a94dfb0247ef1e2f0c955f84aa749b84eb330f0e43a1043b93aa5741c",
  "anchors_sha": "0b331768533b4ab15b6075998e6a1d06013500363fbf8cfa764565b42b157a37"
 },
 "short": {
  "paragraph": "Nineteen public-domain Chopin recordings, four minutes and forty-nine seconds of training on a five-year-old graphics card, and a dial you can hear - five clips on the page that differ from each other in exactly one way.",
  "from_dek": true,
  "counts": {
   "words": 2290,
   "minutes": 10,
   "tables": 0,
   "kit": true
  },
  "bullets": [
   {
    "text": "the model has 2.4 billion",
    "figure": "2.4 billion",
    "cite": "a-music-model-you-can-run-at-home",
    "quote": "It is open-weights under the MIT licence, which means the model itself is free to download and run, all 2.4 billion parameters of it - the adjustable numbers inside it - on one ordinary graphics card in your own machine.",
    "span": [
     1494,
     1506
    ]
   },
   {
    "text": "the adapter adds about 44 million",
    "figure": "44 million",
    "cite": "the-trick-a-small-adapter-not-a-new-model",
    "quote": "It adds about 44 million parameters of its own, under two percent of the model's.",
    "span": [
     3505,
     3516
    ]
   },
   {
    "text": "the training run took 4:51",
    "figure": "4:51",
    "cite": "five-minutes-of-training-honestly-accounted",
    "quote": "The recordings go through four at a time, so each pass makes five weight updates - fifty in all, a startlingly small number of times to change a model's mind about anything - in 4 minutes 49 seconds by the trainer's own clock (4:51 counting loading the model and saving the adapter).",
    "span": [
     6463,
     6467
    ]
   }
  ]
 },
 "sections": [
  {
   "id": "a-music-model-you-can-run-at-home",
   "heading": "A music model you can run at home",
   "level": 2,
   "span": [
    1055,
    1996
   ],
   "chunks": [
    [
     1055,
     1996
    ]
   ],
   "chars": 941,
   "digest": "ACE-Step 1.5 is an open-weights music generator with 2.4 billion parameters that runs on one ordinary graphics card. It produces stereo audio at 48 kHz from short descriptions and is available in a fast variant and a slower one that takes roughly six times as many rendering steps.",
   "digest_skipped": null
  },
  {
   "id": "what-it-sounds-like-on-its-own",
   "heading": "What it sounds like on its own",
   "level": 2,
   "span": [
    1996,
    2761
   ],
   "chunks": [
    [
     1996,
     2761
    ]
   ],
   "chars": 765,
   "digest": "The model's baseline performance is demonstrated using a sixty second request for solo piano. This baseline, which uses one sentence of request and no Chopin, serves as the floor against which all other clips on the page are measured.",
   "digest_skipped": null
  },
  {
   "id": "the-trick-a-small-adapter-not-a-new-model",
   "heading": "The trick: a small adapter, not a new model",
   "level": 2,
   "span": [
    2761,
    4490
   ],
   "chunks": [
    [
     2761,
     4490
    ]
   ],
   "chars": 1729,
   "digest": "Instead of retraining the entire model, a small 88-megabyte adapter is used via LoRA. This adapter adds about 44 million parameters, which is under two percent of the model's total. The adapter's strength can be adjusted using a dial from zero to full.",
   "digest_skipped": null
  },
  {
   "id": "five-minutes-of-training-honestly-accounted",
   "heading": "Five minutes of training, honestly accounted",
   "level": 2,
   "span": [
    4490,
    7892
   ],
   "chunks": [
    [
     4490,
     7892
    ]
   ],
   "chars": 3402,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-dial",
   "heading": "The dial",
   "level": 2,
   "span": [
    7892,
    9907
   ],
   "chunks": [
    [
     7892,
     9907
    ]
   ],
   "chars": 2015,
   "digest": "Five clips demonstrate the effect of the adapter's strength, ranging from 0% to full strength. While a crude ruler shows each step moves the audio measurably further from the baseline, no formal, blinded listening test has been run to determine if the changes sound like Chopin.",
   "digest_skipped": null
  },
  {
   "id": "what-we-did-not-measure",
   "heading": "What we did NOT measure",
   "level": 2,
   "span": [
    9907,
    10492
   ],
   "chunks": [
    [
     9907,
     10492
    ]
   ],
   "chars": 585,
   "digest": "The article notes that it has not measured whether clips sound like Chopin, whether the requested tempo and key were honoured, if a smaller GPU could have run the training, or what happens past ten passes.",
   "digest_skipped": null
  },
  {
   "id": "what-to-take-with-you",
   "heading": "What to take with you",
   "level": 2,
   "span": [
    10492,
    11363
   ],
   "chunks": [
    [
     10492,
     11363
    ]
   ],
   "chars": 871,
   "digest": "The article concludes that a style adapter is a small, detachable 88-megabyte file. Teaching it one composer took under five minutes on a consumer graphics card using mechanical captions, and while the dial provably changes the sound, it has not been ruled to be Chopin.",
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work",
   "heading": "How to check our work",
   "level": 2,
   "span": [
    11363,
    11900
   ],
   "chunks": [
    [
     11363,
     11900
    ]
   ],
   "chars": 537,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "who-ran-this-and-thanks",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    11900,
    13025
   ],
   "chunks": [
    [
     11900,
     13025
    ]
   ],
   "chars": 1125,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "the-rest-of-the-seminar",
   "heading": "The rest of the seminar",
   "level": 2,
   "span": [
    13025,
    14049
   ],
   "chunks": [
    [
     13025,
     14049
    ]
   ],
   "chars": 1024,
   "digest": null,
   "digest_skipped": "credits"
  }
 ],
 "chips": [
  {
   "id": "c-5da86616",
   "q": "how does the ace-step 1.5 music generator function?",
   "a": "The model works similarly to image generators by taking a short text description, optionally including lyrics, and producing finished stereo audio at 48 kHz. It is an open-weights model with 2.4 billion parameters that can be run on a single graphics card without an account or cloud service.",
   "cites": [
    "a-music-model-you-can-run-at-home"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-b52009f6",
   "q": "what defines the baseline audio for this experiment?",
   "a": "The baseline is a sixty-second clip of solo piano generated from a specific verbatim request for a romantic era nocturne with expressive rubato. This clip, produced with no additional training data, serves as the floor against which all other clips are measured.",
   "cites": [
    "what-it-sounds-like-on-its-own"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-53de9107",
   "q": "how does a lora adapter interact with the base model?",
   "a": "A LoRA adapter acts like a lens on a camera; it is a small, detachable file that leaves the large model frozen while nudging its behavior. This specific adapter is 88 megabytes and adds approximately 44 million parameters, which is less than two percent of the original model.",
   "cites": [
    "the-trick-a-small-adapter-not-a-new-model"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-df5c8682",
   "q": "what was the duration and scale of the training process?",
   "a": "The training process took 4 minutes 49 seconds to complete ten passes over nineteen recordings. It involved fifty weight updates and reached a peak memory usage of 5.7 gigabytes according to the trainer's bookkeeping.",
   "cites": [
    "five-minutes-of-training-honestly-accounted"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-a77f46bf",
   "q": "how does the strength of the adapter affect the output?",
   "a": "The adapter's strength functions like a dial from zero to full. As the dial increases, the audio moves measurably further from the baseline in a strict, non-backtracking order, though the dial only measures change rather than specific musical accuracy.",
   "cites": [
    "the-dial"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-ec000eb2",
   "q": "what aspects of the audio were left untested?",
   "a": "The experiment did not include a formal, blinded listening test to determine if the clips actually sound like Chopin. It also did not verify if the specific tempo and key requested were honored, nor did it test the effects of more than ten training passes.",
   "cites": [
    "what-we-did-not-measure"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-2e09ae4e",
   "q": "what are the primary takeaways regarding style adapters?",
   "a": "Style adapters are small, detachable files that do not alter the underlying model. Training a model on a single composer can be achieved in under five minutes using mechanical captions and minimal weight updates, making the cost of experimentation very low.",
   "cites": [
    "what-to-take-with-you"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "half-an-hour-with-dead-composers",
   "why": "shares ground with § What to take with you · § Previously, in five minutes"
  },
  {
   "slug": "ten-minutes-with-living-artists",
   "why": "shares ground with § The dial · § Twelve requests, four ways"
  },
  {
   "slug": "listen-for-yourself",
   "why": "shares ground with § The dial · § What to listen for"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nThe recordings that taught the adapter: Musopen's *Set Chopin Free* project, dedicated\nto the public domain under CC0 — a dedication that demands no credit, which is exactly\nwhy we give it. The model: ACE-Step 1.5 (MIT), whose model card asks users to clearly\ndisclose AI involvement — every clip on this page is AI-generated, and says so. The\ntrainer: Side-Step (github.com/koda-dernet/Side-Step), the corrected-timestep LoRA\ntrainer vendored in ACE-Step 1.5; its upstream licence is CC BY-NC-SA 4.0, though the\nvendored copy says it follows ACE-Step's MIT — we proceed on the stricter reading, and\nwrote to ask.\n\nAnd beneath all of it, the open tools this work stood on without modifying: PyTorch and\ntorchaudio, which ran the training; Hugging Face's PEFT library, which is the LoRA\nmachinery itself; FFmpeg, the decoder that read every training file; and the Internet\nArchive's Wayback Machine, which held the licence page the living web refused us. None\nof them owed us anything. A small team and a fleet of AI agents did the work; the\nhumans signed the numbers.",
 "kit": {
  "url": "https://research.strata2signal.com/chopin-in-five-minutes/data/",
  "licence": "CC BY 4.0",
  "files": [
   "README.md",
   "adapter-config.json",
   "clips.json",
   "corpus-manifest.tsv",
   "render-request.json",
   "rights-chain.md",
   "train.log"
  ]
 },
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 3,
  "digests_written": 7,
  "digests_kept": 6,
  "chips_written": 7,
  "chips_kept": 7,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 0,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 1,
   "profanity": 0
  },
  "new_noun_tokens_checked": [
   "0%",
   "0.61",
   "0.94",
   "1.5",
   "2.4",
   "3090",
   "4",
   "44",
   "48",
   "49",
   "5.7",
   "51",
   "88",
   "88-megabyte",
   "Chopin",
   "GPU",
   "LoRA"
  ],
  "new_noun_tokens_withheld": 1,
  "figure_definition": "v3",
  "drops": [
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 230,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   }
  ]
 },
 "generated_utc": "2026-09-09T01:25:51Z"
}