initial deploy

.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+*.pyc
+.DS_Store
+.venv/
+MMT-repo/
+_outputs/

README.md

@@ -1,13 +1,21 @@
 ---
 title: MMT
-emoji: 🏆
+emoji: 🛤️
 colorFrom: blue
 colorTo: indigo
 sdk: gradio
-sdk_version: 6.20.0
-python_version: '3.13'
+sdk_version: 5.28.0
+python_version: '3.11'
 app_file: app.py
 pinned: false
+license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Multitrack Music Transformer (HARP)
+
+HARP wrapper around the official implementation of "Multitrack Music Transformer" (ICASSP 2023).
+Generates symbolic multi-instrument music — optionally continuing from an uploaded MIDI seed, or
+unconditioned if none is given.
+
+- Paper: https://arxiv.org/abs/2207.06983
+- Code: https://github.com/salu133445/mmt

SOURCES.md

@@ -0,0 +1,5 @@
+# Sources — MMT
+
+- Source repo: https://github.com/salu133445/mmt.git
+- Paper: https://arxiv.org/abs/2207.06983
+- Hardware: undetermined, started on cpu-basic — run `probe_space.py --resize` once app.py is deployed to decide gpu vs cpu

app.py

@@ -0,0 +1,209 @@
+import sys
+
+sys.stdout.reconfigure(line_buffering=True)
+
+try:
+    import spaces
+except ImportError:
+    # keep @spaces.GPU usable as a no-op; ZeroGPU requires this exact name.
+    class spaces:
+        class GPU:
+            def __init__(self, func=None, duration=60):
+                self.func = func
+
+            def __call__(self, *args, **kwargs):
+                if self.func is not None:
+                    return self.func(*args, **kwargs)
+                func = args[0]
+                return func
+
+
+from pathlib import Path
+
+# mmt/ keeps the original repo's bare intra-package imports (`import utils`,
+# `import representation`), so it needs to be on sys.path directly rather
+# than imported as a "mmt.*" package.
+sys.path.insert(0, str(Path(__file__).parent / "mmt"))
+
+import gradio as gr
+import muspy
+import numpy as np
+import torch
+from pyharp import ModelCard, build_endpoint, get_default_path
+
+import music_x_transformers
+import representation
+import utils as mmt_utils
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+CKPT_DIR = Path(__file__).parent / "checkpoint"
+
+train_args = mmt_utils.load_json(CKPT_DIR / "train-args.json")
+encoding = representation.get_encoding()
+
+
+def load_model():
+    """Build the model on CPU and load the sod/ape checkpoint weights. Stays on
+    CPU here even if GPU hardware is available -- ZeroGPU only allows touching
+    CUDA from inside an @spaces.GPU call, not at module load time."""
+    m = music_x_transformers.MusicXTransformer(
+        dim=train_args["dim"],
+        encoding=encoding,
+        depth=train_args["layers"],
+        heads=train_args["heads"],
+        max_seq_len=train_args["max_seq_len"],
+        max_beat=train_args["max_beat"],
+        rotary_pos_emb=train_args["rel_pos_emb"],
+        use_abs_pos_emb=train_args["abs_pos_emb"],
+        emb_dropout=train_args["dropout"],
+        attn_dropout=train_args["dropout"],
+        ff_dropout=train_args["dropout"],
+    )
+    state_dict = torch.load(
+        CKPT_DIR / "checkpoints" / "best_model.pt", map_location="cpu"
+    )
+    m.load_state_dict(state_dict)
+    m.eval()
+    return m
+
+
+# Checkpoint is small (~80MB) and lives in the Space repo, so load it
+# synchronously at startup rather than in a background thread.
+model = load_model()
+model_ready = False  # has the model been moved onto DEVICE yet?
+
+# Fetch upfront so failures show in logs, not after a request.
+muspy.download_musescore_soundfont()
+
+
+def encode_seed(midi_path, encoding, max_beat):
+    """Turn an uploaded MIDI file into the token prefix the model continues
+    from. Mirrors mmt/convert_sod.py's resolution-adjustment step and
+    mmt/dataset.py's max_beat trim, since generate.py normally gets both for
+    free from the preprocessed dataset it reads conditioning prefixes from."""
+    music = muspy.read(midi_path)
+    music.adjust_resolution(encoding["resolution"])
+    for track in music:
+        for note in track:
+            if note.duration == 0:
+                note.duration = 1
+    music.remove_duplicate()
+
+    notes = representation.extract_notes(music, encoding["resolution"])
+    if len(notes) == 0:
+        raise gr.Error("No usable notes found in the uploaded MIDI (check instruments are General MIDI, non-drum).")
+    n_beats = notes[-1, 0] + 1
+    if n_beats > max_beat:
+        notes = notes[notes[:, 0] < max_beat]
+
+    codes = representation.encode_notes(notes, encoding)
+    return codes[:-1]  # drop the trailing EOS row so generation continues instead of stopping immediately
+
+
+@spaces.GPU
+@torch.inference_mode()
+def process_fn(input_midi_path, generation_length, temperature, focus):
+    """Generate a multi-instrument continuation of an uploaded MIDI seed, or an
+    unconditioned sample if no seed is given, and return it as MIDI + a rendered
+    audio preview."""
+    global model, model_ready
+    if not model_ready:
+        model = model.to(DEVICE)
+        model_ready = True
+
+    sos = encoding["type_code_map"]["start-of-song"]
+    eos = encoding["type_code_map"]["end-of-song"]
+
+    if input_midi_path:
+        prefix_codes = encode_seed(input_midi_path, encoding, train_args["max_beat"])
+        tgt_start = torch.tensor(prefix_codes, dtype=torch.long, device=DEVICE).unsqueeze(0)
+    else:
+        tgt_start = torch.zeros((1, 1, 6), dtype=torch.long, device=DEVICE)
+        tgt_start[:, 0, 0] = sos
+
+    generated = model.generate(
+        tgt_start,
+        int(generation_length),
+        eos_token=eos,
+        temperature=temperature,
+        filter_logits_fn="top_k",
+        filter_thres=focus,
+        monotonicity_dim=("type", "beat"),
+    )
+    full_codes = torch.cat((tgt_start, generated), 1)[0].cpu().numpy()
+
+    music = representation.decode(full_codes, encoding)
+
+    midi_path = get_default_path(ext=".mid")
+    music.write(midi_path)
+
+    audio_path = get_default_path(ext=".wav")
+    music.write(audio_path)  # uses the MuseScore General soundfont muspy fetches on first use
+
+    return midi_path, audio_path
+
+
+model_card = ModelCard(
+    name="Multitrack Music Transformer",
+    description=(
+        "Generates symbolic multi-instrument music. Upload a MIDI seed to continue it, "
+        "or leave it empty for an unconditioned generation. Trained on the Symbolic "
+        "Orchestral Database (SOD)."
+    ),
+    author="Hao-Wen Dong, Ke Chen, Shlomo Dubnov, Julian McAuley, Taylor Berg-Kirkpatrick",
+    tags=["symbolic-music", "midi", "multitrack", "transformer"],
+)
+
+
+with gr.Blocks() as demo:
+    input_components = [
+        gr.File(
+            type="filepath",
+            label="Seed MIDI (optional)",
+            file_types=[".mid", ".midi"],
+        ).harp_required(False),
+        gr.Slider(
+            minimum=64,
+            maximum=1024,
+            step=64,
+            value=1024,
+            label="Generation Length",
+            info="Number of musical events to generate (default: 1024, per repo config)",
+        ),
+        gr.Slider(
+            minimum=0.1,
+            maximum=2.0,
+            step=0.1,
+            value=1.0,
+            label="Randomness",
+            info="Sampling temperature -- higher wanders more (default: 1.0, per repo config)",
+        ),
+        gr.Slider(
+            minimum=0.5,
+            maximum=0.99,
+            step=0.01,
+            value=0.9,
+            label="Focus",
+            info="Higher keeps only the most likely notes; lower allows more variety (default: 0.9, per repo config)",
+        ),
+    ]
+    output_components = [
+        gr.File(
+            label="Generated MIDI",
+            file_types=[".mid", ".midi"],
+        ).set_info("The generated multi-instrument MIDI."),
+        gr.Audio(
+            type="filepath",
+            label="Audio Preview",
+        ).set_info("A synthesized audio preview of the generated MIDI."),
+    ]
+
+    build_endpoint(
+        model_card=model_card,
+        input_components=input_components,
+        output_components=output_components,
+        process_fn=process_fn,
+    )
+
+if __name__ == "__main__":
+    demo.queue().launch(pwa=True)

checkpoint/checkpoints/best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c24d1ba5b4e6dbeb346b24734c0f7d0b15f3d3c417490f6d106f49d2f000673
+size 79821244

checkpoint/train-args.json

@@ -0,0 +1 @@
+{"dataset": "sod", "train_names": "data/sod/processed/train-names.txt", "valid_names": "data/sod/processed/valid-names.txt", "in_dir": "data/sod/processed/notes", "out_dir": "/data/herman/workspace/mtmt/exp/sod/ape", "batch_size": 8, "use_csv": false, "aug": true, "max_seq_len": 1024, "max_beat": 256, "dim": 512, "layers": 6, "heads": 8, "dropout": 0.2, "abs_pos_emb": true, "rel_pos_emb": false, "steps": 200000, "valid_steps": 1000, "early_stopping": true, "early_stopping_tolerance": 20, "learning_rate": 0.0005, "lr_warmup_steps": 5000, "lr_decay_steps": 100000, "lr_decay_multiplier": 0.1, "grad_norm_clip": 1.0, "gpu": 7, "jobs": 8, "quiet": false}

mmt/__init__.py

@@ -0,0 +1 @@
+"""MMT model."""

mmt/encoding.json

@@ -0,0 +1 @@
+{"resolution": 24, "max_beat": 1024, "max_duration": 384, "dimensions": ["type", "beat", "position", "pitch", "duration", "instrument"], "n_tokens": [5, 1026, 25, 129, 25, 65], "type_code_map": {"start-of-song": 0, "instrument": 1, "start-of-notes": 2, "note": 3, "end-of-song": 4}, "beat_code_map": {"0": 1, "1": 2, "2": 3, "3": 4, "4": 5, "5": 6, "6": 7, "7": 8, "8": 9, "9": 10, "10": 11, "11": 12, "12": 13, "13": 14, "14": 15, "15": 16, "16": 17, "17": 18, "18": 19, "19": 20, "20": 21, "21": 22, "22": 23, "23": 24, "24": 25, "25": 26, "26": 27, "27": 28, "28": 29, "29": 30, "30": 31, "31": 32, "32": 33, "33": 34, "34": 35, "35": 36, "36": 37, "37": 38, "38": 39, "39": 40, "40": 41, "41": 42, "42": 43, "43": 44, "44": 45, "45": 46, "46": 47, "47": 48, "48": 49, "49": 50, "50": 51, "51": 52, "52": 53, "53": 54, "54": 55, "55": 56, "56": 57, "57": 58, "58": 59, "59": 60, "60": 61, "61": 62, "62": 63, "63": 64, "64": 65, "65": 66, "66": 67, "67": 68, "68": 69, "69": 70, "70": 71, "71": 72, "72": 73, "73": 74, "74": 75, "75": 76, "76": 77, "77": 78, "78": 79, "79": 80, "80": 81, "81": 82, "82": 83, "83": 84, "84": 85, "85": 86, "86": 87, "87": 88, "88": 89, "89": 90, "90": 91, "91": 92, "92": 93, "93": 94, "94": 95, "95": 96, "96": 97, "97": 98, "98": 99, "99": 100, "100": 101, "101": 102, "102": 103, "103": 104, "104": 105, "105": 106, "106": 107, "107": 108, "108": 109, "109": 110, "110": 111, "111": 112, "112": 113, "113": 114, "114": 115, "115": 116, "116": 117, "117": 118, "118": 119, "119": 120, "120": 121, "121": 122, "122": 123, "123": 124, "124": 125, "125": 126, "126": 127, "127": 128, "128": 129, "129": 130, "130": 131, "131": 132, "132": 133, "133": 134, "134": 135, "135": 136, "136": 137, "137": 138, "138": 139, "139": 140, "140": 141, "141": 142, "142": 143, "143": 144, "144": 145, "145": 146, "146": 147, "147": 148, "148": 149, "149": 150, "150": 151, "151": 152, "152": 153, "153": 154, "154": 155, "155": 156, "156": 157, "157": 158, "158": 159, "159": 160, "160": 161, "161": 162, "162": 163, "163": 164, "164": 165, "165": 166, "166": 167, "167": 168, "168": 169, "169": 170, "170": 171, "171": 172, "172": 173, "173": 174, "174": 175, "175": 176, "176": 177, "177": 178, "178": 179, "179": 180, "180": 181, "181": 182, "182": 183, "183": 184, "184": 185, "185": 186, "186": 187, "187": 188, "188": 189, "189": 190, "190": 191, "191": 192, "192": 193, "193": 194, "194": 195, "195": 196, "196": 197, "197": 198, "198": 199, "199": 200, "200": 201, "201": 202, "202": 203, "203": 204, "204": 205, "205": 206, "206": 207, "207": 208, "208": 209, "209": 210, "210": 211, "211": 212, "212": 213, "213": 214, "214": 215, "215": 216, "216": 217, "217": 218, "218": 219, "219": 220, "220": 221, "221": 222, "222": 223, "223": 224, "224": 225, "225": 226, "226": 227, "227": 228, "228": 229, "229": 230, "230": 231, "231": 232, "232": 233, "233": 234, "234": 235, "235": 236, "236": 237, "237": 238, "238": 239, "239": 240, "240": 241, "241": 242, "242": 243, "243": 244, "244": 245, "245": 246, "246": 247, "247": 248, "248": 249, "249": 250, "250": 251, "251": 252, "252": 253, "253": 254, "254": 255, "255": 256, "256": 257, "257": 258, "258": 259, "259": 260, "260": 261, "261": 262, "262": 263, "263": 264, "264": 265, "265": 266, "266": 267, "267": 268, "268": 269, "269": 270, "270": 271, "271": 272, "272": 273, "273": 274, "274": 275, "275": 276, "276": 277, "277": 278, "278": 279, "279": 280, "280": 281, "281": 282, "282": 283, "283": 284, "284": 285, "285": 286, "286": 287, "287": 288, "288": 289, "289": 290, "290": 291, "291": 292, "292": 293, "293": 294, "294": 295, "295": 296, "296": 297, "297": 298, "298": 299, "299": 300, "300": 301, "301": 302, "302": 303, "303": 304, "304": 305, "305": 306, "306": 307, "307": 308, "308": 309, "309": 310, "310": 311, "311": 312, "312": 313, "313": 314, "314": 315, "315": 316, "316": 317, "317": 318, "318": 319, "319": 320, "320": 321, "321": 322, "322": 323, "323": 324, "324": 325, "325": 326, "326": 327, "327": 328, "328": 329, "329": 330, "330": 331, "331": 332, "332": 333, "333": 334, "334": 335, "335": 336, "336": 337, "337": 338, "338": 339, "339": 340, "340": 341, "341": 342, "342": 343, "343": 344, "344": 345, "345": 346, "346": 347, "347": 348, "348": 349, "349": 350, "350": 351, "351": 352, "352": 353, "353": 354, "354": 355, "355": 356, "356": 357, "357": 358, "358": 359, "359": 360, "360": 361, "361": 362, "362": 363, "363": 364, "364": 365, "365": 366, "366": 367, "367": 368, "368": 369, "369": 370, "370": 371, "371": 372, "372": 373, "373": 374, "374": 375, "375": 376, "376": 377, "377": 378, "378": 379, "379": 380, "380": 381, "381": 382, "382": 383, "383": 384, "384": 385, "385": 386, "386": 387, "387": 388, "388": 389, "389": 390, "390": 391, "391": 392, "392": 393, "393": 394, "394": 395, "395": 396, "396": 397, "397": 398, "398": 399, "399": 400, "400": 401, "401": 402, "402": 403, "403": 404, "404": 405, "405": 406, "406": 407, "407": 408, "408": 409, "409": 410, "410": 411, "411": 412, "412": 413, "413": 414, "414": 415, "415": 416, "416": 417, "417": 418, "418": 419, "419": 420, "420": 421, "421": 422, "422": 423, "423": 424, "424": 425, "425": 426, "426": 427, "427": 428, "428": 429, "429": 430, "430": 431, "431": 432, "432": 433, "433": 434, "434": 435, "435": 436, "436": 437, "437": 438, "438": 439, "439": 440, "440": 441, "441": 442, "442": 443, "443": 444, "444": 445, "445": 446, "446": 447, "447": 448, "448": 449, "449": 450, "450": 451, "451": 452, "452": 453, "453": 454, "454": 455, "455": 456, "456": 457, "457": 458, "458": 459, "459": 460, "460": 461, "461": 462, "462": 463, "463": 464, "464": 465, "465": 466, "466": 467, "467": 468, "468": 469, "469": 470, "470": 471, "471": 472, "472": 473, "473": 474, "474": 475, "475": 476, "476": 477, "477": 478, "478": 479, "479": 480, "480": 481, "481": 482, "482": 483, "483": 484, "484": 485, "485": 486, "486": 487, "487": 488, "488": 489, "489": 490, "490": 491, "491": 492, "492": 493, "493": 494, "494": 495, "495": 496, "496": 497, "497": 498, "498": 499, "499": 500, "500": 501, "501": 502, "502": 503, "503": 504, "504": 505, "505": 506, "506": 507, "507": 508, "508": 509, "509": 510, "510": 511, "511": 512, "512": 513, "513": 514, "514": 515, "515": 516, "516": 517, "517": 518, "518": 519, "519": 520, "520": 521, "521": 522, "522": 523, "523": 524, "524": 525, "525": 526, "526": 527, "527": 528, "528": 529, "529": 530, "530": 531, "531": 532, "532": 533, "533": 534, "534": 535, "535": 536, "536": 537, "537": 538, "538": 539, "539": 540, "540": 541, "541": 542, "542": 543, "543": 544, "544": 545, "545": 546, "546": 547, "547": 548, "548": 549, "549": 550, "550": 551, "551": 552, "552": 553, "553": 554, "554": 555, "555": 556, "556": 557, "557": 558, "558": 559, "559": 560, "560": 561, "561": 562, "562": 563, "563": 564, "564": 565, "565": 566, "566": 567, "567": 568, "568": 569, "569": 570, "570": 571, "571": 572, "572": 573, "573": 574, "574": 575, "575": 576, "576": 577, "577": 578, "578": 579, "579": 580, "580": 581, "581": 582, "582": 583, "583": 584, "584": 585, "585": 586, "586": 587, "587": 588, "588": 589, "589": 590, "590": 591, "591": 592, "592": 593, "593": 594, "594": 595, "595": 596, "596": 597, "597": 598, "598": 599, "599": 600, "600": 601, "601": 602, "602": 603, "603": 604, "604": 605, "605": 606, "606": 607, "607": 608, "608": 609, "609": 610, "610": 611, "611": 612, "612": 613, "613": 614, "614": 615, "615": 616, "616": 617, "617": 618, "618": 619, "619": 620, "620": 621, "621": 622, "622": 623, "623": 624, "624": 625, "625": 626, "626": 627, "627": 628, "628": 629, "629": 630, "630": 631, "631": 632, "632": 633, "633": 634, "634": 635, "635": 636, "636": 637, "637": 638, "638": 639, "639": 640, "640": 641, "641": 642, "642": 643, "643": 644, "644": 645, "645": 646, "646": 647, "647": 648, "648": 649, "649": 650, "650": 651, "651": 652, "652": 653, "653": 654, "654": 655, "655": 656, "656": 657, "657": 658, "658": 659, "659": 660, "660": 661, "661": 662, "662": 663, "663": 664, "664": 665, "665": 666, "666": 667, "667": 668, "668": 669, "669": 670, "670": 671, "671": 672, "672": 673, "673": 674, "674": 675, "675": 676, "676": 677, "677": 678, "678": 679, "679": 680, "680": 681, "681": 682, "682": 683, "683": 684, "684": 685, "685": 686, "686": 687, "687": 688, "688": 689, "689": 690, "690": 691, "691": 692, "692": 693, "693": 694, "694": 695, "695": 696, "696": 697, "697": 698, "698": 699, "699": 700, "700": 701, "701": 702, "702": 703, "703": 704, "704": 705, "705": 706, "706": 707, "707": 708, "708": 709, "709": 710, "710": 711, "711": 712, "712": 713, "713": 714, "714": 715, "715": 716, "716": 717, "717": 718, "718": 719, "719": 720, "720": 721, "721": 722, "722": 723, "723": 724, "724": 725, "725": 726, "726": 727, "727": 728, "728": 729, "729": 730, "730": 731, "731": 732, "732": 733, "733": 734, "734": 735, "735": 736, "736": 737, "737": 738, "738": 739, "739": 740, "740": 741, "741": 742, "742": 743, "743": 744, "744": 745, "745": 746, "746": 747, "747": 748, "748": 749, "749": 750, "750": 751, "751": 752, "752": 753, "753": 754, "754": 755, "755": 756, "756": 757, "757": 758, "758": 759, "759": 760, "760": 761, "761": 762, "762": 763, "763": 764, "764": 765, "765": 766, "766": 767, "767": 768, "768": 769, "769": 770, "770": 771, "771": 772, "772": 773, "773": 774, "774": 775, "775": 776, "776": 777, "777": 778, "778": 779, "779": 780, "780": 781, "781": 782, "782": 783, "783": 784, "784": 785, "785": 786, "786": 787, "787": 788, "788": 789, "789": 790, "790": 791, "791": 792, "792": 793, "793": 794, "794": 795, "795": 796, "796": 797, "797": 798, "798": 799, "799": 800, "800": 801, "801": 802, "802": 803, "803": 804, "804": 805, "805": 806, "806": 807, "807": 808, "808": 809, "809": 810, "810": 811, "811": 812, "812": 813, "813": 814, "814": 815, "815": 816, "816": 817, "817": 818, "818": 819, "819": 820, "820": 821, "821": 822, "822": 823, "823": 824, "824": 825, "825": 826, "826": 827, "827": 828, "828": 829, "829": 830, "830": 831, "831": 832, "832": 833, "833": 834, "834": 835, "835": 836, "836": 837, "837": 838, "838": 839, "839": 840, "840": 841, "841": 842, "842": 843, "843": 844, "844": 845, "845": 846, "846": 847, "847": 848, "848": 849, "849": 850, "850": 851, "851": 852, "852": 853, "853": 854, "854": 855, "855": 856, "856": 857, "857": 858, "858": 859, "859": 860, "860": 861, "861": 862, "862": 863, "863": 864, "864": 865, "865": 866, "866": 867, "867": 868, "868": 869, "869": 870, "870": 871, "871": 872, "872": 873, "873": 874, "874": 875, "875": 876, "876": 877, "877": 878, "878": 879, "879": 880, "880": 881, "881": 882, "882": 883, "883": 884, "884": 885, "885": 886, "886": 887, "887": 888, "888": 889, "889": 890, "890": 891, "891": 892, "892": 893, "893": 894, "894": 895, "895": 896, "896": 897, "897": 898, "898": 899, "899": 900, "900": 901, "901": 902, "902": 903, "903": 904, "904": 905, "905": 906, "906": 907, "907": 908, "908": 909, "909": 910, "910": 911, "911": 912, "912": 913, "913": 914, "914": 915, "915": 916, "916": 917, "917": 918, "918": 919, "919": 920, "920": 921, "921": 922, "922": 923, "923": 924, "924": 925, "925": 926, "926": 927, "927": 928, "928": 929, "929": 930, "930": 931, "931": 932, "932": 933, "933": 934, "934": 935, "935": 936, "936": 937, "937": 938, "938": 939, "939": 940, "940": 941, "941": 942, "942": 943, "943": 944, "944": 945, "945": 946, "946": 947, "947": 948, "948": 949, "949": 950, "950": 951, "951": 952, "952": 953, "953": 954, "954": 955, "955": 956, "956": 957, "957": 958, "958": 959, "959": 960, "960": 961, "961": 962, "962": 963, "963": 964, "964": 965, "965": 966, "966": 967, "967": 968, "968": 969, "969": 970, "970": 971, "971": 972, "972": 973, "973": 974, "974": 975, "975": 976, "976": 977, "977": 978, "978": 979, "979": 980, "980": 981, "981": 982, "982": 983, "983": 984, "984": 985, "985": 986, "986": 987, "987": 988, "988": 989, "989": 990, "990": 991, "991": 992, "992": 993, "993": 994, "994": 995, "995": 996, "996": 997, "997": 998, "998": 999, "999": 1000, "1000": 1001, "1001": 1002, "1002": 1003, "1003": 1004, "1004": 1005, "1005": 1006, "1006": 1007, "1007": 1008, "1008": 1009, "1009": 1010, "1010": 1011, "1011": 1012, "1012": 1013, "1013": 1014, "1014": 1015, "1015": 1016, "1016": 1017, "1017": 1018, "1018": 1019, "1019": 1020, "1020": 1021, "1021": 1022, "1022": 1023, "1023": 1024, "1024": 1025, "null": 0}, "position_code_map": {"0": 1, "1": 2, "2": 3, "3": 4, "4": 5, "5": 6, "6": 7, "7": 8, "8": 9, "9": 10, "10": 11, "11": 12, "12": 13, "13": 14, "14": 15, "15": 16, "16": 17, "17": 18, "18": 19, "19": 20, "20": 21, "21": 22, "22": 23, "23": 24, "null": 0}, "pitch_code_map": {"0": 1, "1": 2, "2": 3, "3": 4, "4": 5, "5": 6, "6": 7, "7": 8, "8": 9, "9": 10, "10": 11, "11": 12, "12": 13, "13": 14, "14": 15, "15": 16, "16": 17, "17": 18, "18": 19, "19": 20, "20": 21, "21": 22, "22": 23, "23": 24, "24": 25, "25": 26, "26": 27, "27": 28, "28": 29, "29": 30, "30": 31, "31": 32, "32": 33, "33": 34, "34": 35, "35": 36, "36": 37, "37": 38, "38": 39, "39": 40, "40": 41, "41": 42, "42": 43, "43": 44, "44": 45, "45": 46, "46": 47, "47": 48, "48": 49, "49": 50, "50": 51, "51": 52, "52": 53, "53": 54, "54": 55, "55": 56, "56": 57, "57": 58, "58": 59, "59": 60, "60": 61, "61": 62, "62": 63, "63": 64, "64": 65, "65": 66, "66": 67, "67": 68, "68": 69, "69": 70, "70": 71, "71": 72, "72": 73, "73": 74, "74": 75, "75": 76, "76": 77, "77": 78, "78": 79, "79": 80, "80": 81, "81": 82, "82": 83, "83": 84, "84": 85, "85": 86, "86": 87, "87": 88, "88": 89, "89": 90, "90": 91, "91": 92, "92": 93, "93": 94, "94": 95, "95": 96, "96": 97, "97": 98, "98": 99, "99": 100, "100": 101, "101": 102, "102": 103, "103": 104, "104": 105, "105": 106, "106": 107, "107": 108, "108": 109, "109": 110, "110": 111, "111": 112, "112": 113, "113": 114, "114": 115, "115": 116, "116": 117, "117": 118, "118": 119, "119": 120, "120": 121, "121": 122, "122": 123, "123": 124, "124": 125, "125": 126, "126": 127, "127": 128, "null": 0}, "duration_code_map": {"0": 1, "1": 1, "2": 2, "3": 3, "4": 3, "5": 3, "6": 3, "7": 3, "8": 4, "9": 4, "10": 4, "11": 4, "12": 4, "13": 4, "14": 5, "15": 5, "16": 6, "17": 6, "18": 7, "19": 7, "20": 8, "21": 9, "22": 9, "23": 10, "24": 10, "25": 10, "26": 10, "27": 10, "28": 11, "29": 11, "30": 11, "31": 11, "32": 11, "33": 11, "34": 12, "35": 12, "36": 12, "37": 12, "38": 12, "39": 13, "40": 13, "41": 13, "42": 14, "43": 14, "44": 14, "45": 14, "46": 15, "47": 15, "48": 15, "49": 15, "50": 15, "51": 15, "52": 15, "53": 15, "54": 15, "55": 16, "56": 16, "57": 16, "58": 16, "59": 16, "60": 16, "61": 16, "62": 16, "63": 16, "64": 16, "65": 16, "66": 16, "67": 17, "68": 17, "69": 17, "70": 17, "71": 17, "72": 17, "73": 17, "74": 17, "75": 17, "76": 17, "77": 17, "78": 17, "79": 18, "80": 18, "81": 18, "82": 18, "83": 18, "84": 18, "85": 18, "86": 18, "87": 18, "88": 18, "89": 18, "90": 18, "91": 19, "92": 19, "93": 19, "94": 19, "95": 19, "96": 19, "97": 19, "98": 19, "99": 19, "100": 19, "101": 19, "102": 19, "103": 19, "104": 19, "105": 19, "106": 19, "107": 19, "108": 19, "109": 20, "110": 20, "111": 20, "112": 20, "113": 20, "114": 20, "115": 20, "116": 20, "117": 20, "118": 20, "119": 20, "120": 20, "121": 20, "122": 20, "123": 20, "124": 20, "125": 20, "126": 20, "127": 20, "128": 20, "129": 20, "130": 20, "131": 20, "132": 20, "133": 21, "134": 21, "135": 21, "136": 21, "137": 21, "138": 21, "139": 21, "140": 21, "141": 21, "142": 21, "143": 21, "144": 21, "145": 21, "146": 21, "147": 21, "148": 21, "149": 21, "150": 21, "151": 21, "152": 21, "153": 21, "154": 21, "155": 21, "156": 21, "157": 22, "158": 22, "159": 22, "160": 22, "161": 22, "162": 22, "163": 22, "164": 22, "165": 22, "166": 22, "167": 22, "168": 22, "169": 22, "170": 22, "171": 22, "172": 22, "173": 22, "174": 22, "175": 22, "176": 22, "177": 22, "178": 22, "179": 22, "180": 22, "181": 23, "182": 23, "183": 23, "184": 23, "185": 23, "186": 23, "187": 23, "188": 23, "189": 23, "190": 23, "191": 23, "192": 23, "193": 23, "194": 23, "195": 23, "196": 23, "197": 23, "198": 23, "199": 23, "200": 23, "201": 23, "202": 23, "203": 23, "204": 23, "205": 23, "206": 23, "207": 23, "208": 23, "209": 23, "210": 23, "211": 23, "212": 23, "213": 23, "214": 23, "215": 23, "216": 23, "217": 23, "218": 23, "219": 23, "220": 23, "221": 23, "222": 23, "223": 23, "224": 23, "225": 23, "226": 23, "227": 23, "228": 23, "229": 23, "230": 23, "231": 23, "232": 23, "233": 23, "234": 23, "235": 23, "236": 23, "237": 23, "238": 23, "239": 23, "240": 23, "241": 23, "242": 23, "243": 23, "244": 23, "245": 23, "246": 23, "247": 23, "248": 23, "249": 23, "250": 23, "251": 23, "252": 23, "253": 23, "254": 23, "255": 23, "256": 23, "257": 23, "258": 23, "259": 23, "260": 23, "261": 23, "262": 23, "263": 23, "264": 23, "265": 23, "266": 23, "267": 23, "268": 23, "269": 23, "270": 23, "271": 23, "272": 23, "273": 23, "274": 23, "275": 23, "276": 23, "277": 23, "278": 23, "279": 23, "280": 23, "281": 23, "282": 23, "283": 23, "284": 23, "285": 23, "286": 23, "287": 23, "288": 23, "289": 24, "290": 24, "291": 24, "292": 24, "293": 24, "294": 24, "295": 24, "296": 24, "297": 24, "298": 24, "299": 24, "300": 24, "301": 24, "302": 24, "303": 24, "304": 24, "305": 24, "306": 24, "307": 24, "308": 24, "309": 24, "310": 24, "311": 24, "312": 24, "313": 24, "314": 24, "315": 24, "316": 24, "317": 24, "318": 24, "319": 24, "320": 24, "321": 24, "322": 24, "323": 24, "324": 24, "325": 24, "326": 24, "327": 24, "328": 24, "329": 24, "330": 24, "331": 24, "332": 24, "333": 24, "334": 24, "335": 24, "336": 24, "337": 24, "338": 24, "339": 24, "340": 24, "341": 24, "342": 24, "343": 24, "344": 24, "345": 24, "346": 24, "347": 24, "348": 24, "349": 24, "350": 24, "351": 24, "352": 24, "353": 24, "354": 24, "355": 24, "356": 24, "357": 24, "358": 24, "359": 24, "360": 24, "361": 24, "362": 24, "363": 24, "364": 24, "365": 24, "366": 24, "367": 24, "368": 24, "369": 24, "370": 24, "371": 24, "372": 24, "373": 24, "374": 24, "375": 24, "376": 24, "377": 24, "378": 24, "379": 24, "380": 24, "381": 24, "382": 24, "383": 24, "384": 24, "null": 0}, "instrument_code_map": {"piano": 1, "electric-piano": 2, "harpsichord": 3, "clavinet": 4, "celesta": 5, "glockenspiel": 6, "music-box": 7, "vibraphone": 8, "marimba": 9, "xylophone": 10, "tubular-bells": 11, "dulcimer": 12, "organ": 13, "church-organ": 14, "accordion": 15, "harmonica": 16, "bandoneon": 17, "nylon-string-guitar": 18, "steel-string-guitar": 19, "electric-guitar": 20, "bass": 21, "electric-bass": 22, "slap-bass": 23, "synth-bass": 24, "violin": 25, "viola": 26, "cello": 27, "contrabass": 28, "harp": 29, "timpani": 30, "strings": 31, "synth-strings": 32, "voices": 33, "orchestra-hit": 34, "trumpet": 35, "trombone": 36, "tuba": 37, "horn": 38, "brasses": 39, "synth-brasses": 40, "soprano-saxophone": 41, "alto-saxophone": 42, "tenor-saxophone": 43, "baritone-saxophone": 44, "oboe": 45, "english-horn": 46, "bassoon": 47, "clarinet": 48, "piccolo": 49, "flute": 50, "recorder": 51, "pan-flute": 52, "ocarina": 53, "lead": 54, "pad": 55, "sitar": 56, "banjo": 57, "shamisen": 58, "koto": 59, "kalimba": 60, "bag-pipe": 61, "shehnai": 62, "melodic-tom": 63, "synth-drums": 64, "null": 0}, "code_type_map": {"0": "start-of-song", "1": "instrument", "2": "start-of-notes", "3": "note", "4": "end-of-song"}, "code_beat_map": {"1": 0, "2": 1, "3": 2, "4": 3, "5": 4, "6": 5, "7": 6, "8": 7, "9": 8, "10": 9, "11": 10, "12": 11, "13": 12, "14": 13, "15": 14, "16": 15, "17": 16, "18": 17, "19": 18, "20": 19, "21": 20, "22": 21, "23": 22, "24": 23, "25": 24, "26": 25, "27": 26, "28": 27, "29": 28, "30": 29, "31": 30, "32": 31, "33": 32, "34": 33, "35": 34, "36": 35, "37": 36, "38": 37, "39": 38, "40": 39, "41": 40, "42": 41, "43": 42, "44": 43, "45": 44, "46": 45, "47": 46, "48": 47, "49": 48, "50": 49, "51": 50, "52": 51, "53": 52, "54": 53, "55": 54, "56": 55, "57": 56, "58": 57, "59": 58, "60": 59, "61": 60, "62": 61, "63": 62, "64": 63, "65": 64, "66": 65, "67": 66, "68": 67, "69": 68, "70": 69, "71": 70, "72": 71, "73": 72, "74": 73, "75": 74, "76": 75, "77": 76, "78": 77, "79": 78, "80": 79, "81": 80, "82": 81, "83": 82, "84": 83, "85": 84, "86": 85, "87": 86, "88": 87, "89": 88, "90": 89, "91": 90, "92": 91, "93": 92, "94": 93, "95": 94, "96": 95, "97": 96, "98": 97, "99": 98, "100": 99, "101": 100, "102": 101, "103": 102, "104": 103, "105": 104, "106": 105, "107": 106, "108": 107, "109": 108, "110": 109, "111": 110, "112": 111, "113": 112, "114": 113, "115": 114, "116": 115, "117": 116, "118": 117, "119": 118, "120": 119, "121": 120, "122": 121, "123": 122, "124": 123, "125": 124, "126": 125, "127": 126, "128": 127, "129": 128, "130": 129, "131": 130, "132": 131, "133": 132, "134": 133, "135": 134, "136": 135, "137": 136, "138": 137, "139": 138, "140": 139, "141": 140, "142": 141, "143": 142, "144": 143, "145": 144, "146": 145, "147": 146, "148": 147, "149": 148, "150": 149, "151": 150, "152": 151, "153": 152, "154": 153, "155": 154, "156": 155, "157": 156, "158": 157, "159": 158, "160": 159, "161": 160, "162": 161, "163": 162, "164": 163, "165": 164, "166": 165, "167": 166, "168": 167, "169": 168, "170": 169, "171": 170, "172": 171, "173": 172, "174": 173, "175": 174, "176": 175, "177": 176, "178": 177, "179": 178, "180": 179, "181": 180, "182": 181, "183": 182, "184": 183, "185": 184, "186": 185, "187": 186, "188": 187, "189": 188, "190": 189, "191": 190, "192": 191, "193": 192, "194": 193, "195": 194, "196": 195, "197": 196, "198": 197, "199": 198, "200": 199, "201": 200, "202": 201, "203": 202, "204": 203, "205": 204, "206": 205, "207": 206, "208": 207, "209": 208, "210": 209, "211": 210, "212": 211, "213": 212, "214": 213, "215": 214, "216": 215, "217": 216, "218": 217, "219": 218, "220": 219, "221": 220, "222": 221, "223": 222, "224": 223, "225": 224, "226": 225, "227": 226, "228": 227, "229": 228, "230": 229, "231": 230, "232": 231, "233": 232, "234": 233, "235": 234, "236": 235, "237": 236, "238": 237, "239": 238, "240": 239, "241": 240, "242": 241, "243": 242, "244": 243, "245": 244, "246": 245, "247": 246, "248": 247, "249": 248, "250": 249, "251": 250, "252": 251, "253": 252, "254": 253, "255": 254, "256": 255, "257": 256, "258": 257, "259": 258, "260": 259, "261": 260, "262": 261, "263": 262, "264": 263, "265": 264, "266": 265, "267": 266, "268": 267, "269": 268, "270": 269, "271": 270, "272": 271, "273": 272, "274": 273, "275": 274, "276": 275, "277": 276, "278": 277, "279": 278, "280": 279, "281": 280, "282": 281, "283": 282, "284": 283, "285": 284, "286": 285, "287": 286, "288": 287, "289": 288, "290": 289, "291": 290, "292": 291, "293": 292, "294": 293, "295": 294, "296": 295, "297": 296, "298": 297, "299": 298, "300": 299, "301": 300, "302": 301, "303": 302, "304": 303, "305": 304, "306": 305, "307": 306, "308": 307, "309": 308, "310": 309, "311": 310, "312": 311, "313": 312, "314": 313, "315": 314, "316": 315, "317": 316, "318": 317, "319": 318, "320": 319, "321": 320, "322": 321, "323": 322, "324": 323, "325": 324, "326": 325, "327": 326, "328": 327, "329": 328, "330": 329, "331": 330, "332": 331, "333": 332, "334": 333, "335": 334, "336": 335, "337": 336, "338": 337, "339": 338, "340": 339, "341": 340, "342": 341, "343": 342, "344": 343, "345": 344, "346": 345, "347": 346, "348": 347, "349": 348, "350": 349, "351": 350, "352": 351, "353": 352, "354": 353, "355": 354, "356": 355, "357": 356, "358": 357, "359": 358, "360": 359, "361": 360, "362": 361, "363": 362, "364": 363, "365": 364, "366": 365, "367": 366, "368": 367, "369": 368, "370": 369, "371": 370, "372": 371, "373": 372, "374": 373, "375": 374, "376": 375, "377": 376, "378": 377, "379": 378, "380": 379, "381": 380, "382": 381, "383": 382, "384": 383, "385": 384, "386": 385, "387": 386, "388": 387, "389": 388, "390": 389, "391": 390, "392": 391, "393": 392, "394": 393, "395": 394, "396": 395, "397": 396, "398": 397, "399": 398, "400": 399, "401": 400, "402": 401, "403": 402, "404": 403, "405": 404, "406": 405, "407": 406, "408": 407, "409": 408, "410": 409, "411": 410, "412": 411, "413": 412, "414": 413, "415": 414, "416": 415, "417": 416, "418": 417, "419": 418, "420": 419, "421": 420, "422": 421, "423": 422, "424": 423, "425": 424, "426": 425, "427": 426, "428": 427, "429": 428, "430": 429, "431": 430, "432": 431, "433": 432, "434": 433, "435": 434, "436": 435, "437": 436, "438": 437, "439": 438, "440": 439, "441": 440, "442": 441, "443": 442, "444": 443, "445": 444, "446": 445, "447": 446, "448": 447, "449": 448, "450": 449, "451": 450, "452": 451, "453": 452, "454": 453, "455": 454, "456": 455, "457": 456, "458": 457, "459": 458, "460": 459, "461": 460, "462": 461, "463": 462, "464": 463, "465": 464, "466": 465, "467": 466, "468": 467, "469": 468, "470": 469, "471": 470, "472": 471, "473": 472, "474": 473, "475": 474, "476": 475, "477": 476, "478": 477, "479": 478, "480": 479, "481": 480, "482": 481, "483": 482, "484": 483, "485": 484, "486": 485, "487": 486, "488": 487, "489": 488, "490": 489, "491": 490, "492": 491, "493": 492, "494": 493, "495": 494, "496": 495, "497": 496, "498": 497, "499": 498, "500": 499, "501": 500, "502": 501, "503": 502, "504": 503, "505": 504, "506": 505, "507": 506, "508": 507, "509": 508, "510": 509, "511": 510, "512": 511, "513": 512, "514": 513, "515": 514, "516": 515, "517": 516, "518": 517, "519": 518, "520": 519, "521": 520, "522": 521, "523": 522, "524": 523, "525": 524, "526": 525, "527": 526, "528": 527, "529": 528, "530": 529, "531": 530, "532": 531, "533": 532, "534": 533, "535": 534, "536": 535, "537": 536, "538": 537, "539": 538, "540": 539, "541": 540, "542": 541, "543": 542, "544": 543, "545": 544, "546": 545, "547": 546, "548": 547, "549": 548, "550": 549, "551": 550, "552": 551, "553": 552, "554": 553, "555": 554, "556": 555, "557": 556, "558": 557, "559": 558, "560": 559, "561": 560, "562": 561, "563": 562, "564": 563, "565": 564, "566": 565, "567": 566, "568": 567, "569": 568, "570": 569, "571": 570, "572": 571, "573": 572, "574": 573, "575": 574, "576": 575, "577": 576, "578": 577, "579": 578, "580": 579, "581": 580, "582": 581, "583": 582, "584": 583, "585": 584, "586": 585, "587": 586, "588": 587, "589": 588, "590": 589, "591": 590, "592": 591, "593": 592, "594": 593, "595": 594, "596": 595, "597": 596, "598": 597, "599": 598, "600": 599, "601": 600, "602": 601, "603": 602, "604": 603, "605": 604, "606": 605, "607": 606, "608": 607, "609": 608, "610": 609, "611": 610, "612": 611, "613": 612, "614": 613, "615": 614, "616": 615, "617": 616, "618": 617, "619": 618, "620": 619, "621": 620, "622": 621, "623": 622, "624": 623, "625": 624, "626": 625, "627": 626, "628": 627, "629": 628, "630": 629, "631": 630, "632": 631, "633": 632, "634": 633, "635": 634, "636": 635, "637": 636, "638": 637, "639": 638, "640": 639, "641": 640, "642": 641, "643": 642, "644": 643, "645": 644, "646": 645, "647": 646, "648": 647, "649": 648, "650": 649, "651": 650, "652": 651, "653": 652, "654": 653, "655": 654, "656": 655, "657": 656, "658": 657, "659": 658, "660": 659, "661": 660, "662": 661, "663": 662, "664": 663, "665": 664, "666": 665, "667": 666, "668": 667, "669": 668, "670": 669, "671": 670, "672": 671, "673": 672, "674": 673, "675": 674, "676": 675, "677": 676, "678": 677, "679": 678, "680": 679, "681": 680, "682": 681, "683": 682, "684": 683, "685": 684, "686": 685, "687": 686, "688": 687, "689": 688, "690": 689, "691": 690, "692": 691, "693": 692, "694": 693, "695": 694, "696": 695, "697": 696, "698": 697, "699": 698, "700": 699, "701": 700, "702": 701, "703": 702, "704": 703, "705": 704, "706": 705, "707": 706, "708": 707, "709": 708, "710": 709, "711": 710, "712": 711, "713": 712, "714": 713, "715": 714, "716": 715, "717": 716, "718": 717, "719": 718, "720": 719, "721": 720, "722": 721, "723": 722, "724": 723, "725": 724, "726": 725, "727": 726, "728": 727, "729": 728, "730": 729, "731": 730, "732": 731, "733": 732, "734": 733, "735": 734, "736": 735, "737": 736, "738": 737, "739": 738, "740": 739, "741": 740, "742": 741, "743": 742, "744": 743, "745": 744, "746": 745, "747": 746, "748": 747, "749": 748, "750": 749, "751": 750, "752": 751, "753": 752, "754": 753, "755": 754, "756": 755, "757": 756, "758": 757, "759": 758, "760": 759, "761": 760, "762": 761, "763": 762, "764": 763, "765": 764, "766": 765, "767": 766, "768": 767, "769": 768, "770": 769, "771": 770, "772": 771, "773": 772, "774": 773, "775": 774, "776": 775, "777": 776, "778": 777, "779": 778, "780": 779, "781": 780, "782": 781, "783": 782, "784": 783, "785": 784, "786": 785, "787": 786, "788": 787, "789": 788, "790": 789, "791": 790, "792": 791, "793": 792, "794": 793, "795": 794, "796": 795, "797": 796, "798": 797, "799": 798, "800": 799, "801": 800, "802": 801, "803": 802, "804": 803, "805": 804, "806": 805, "807": 806, "808": 807, "809": 808, "810": 809, "811": 810, "812": 811, "813": 812, "814": 813, "815": 814, "816": 815, "817": 816, "818": 817, "819": 818, "820": 819, "821": 820, "822": 821, "823": 822, "824": 823, "825": 824, "826": 825, "827": 826, "828": 827, "829": 828, "830": 829, "831": 830, "832": 831, "833": 832, "834": 833, "835": 834, "836": 835, "837": 836, "838": 837, "839": 838, "840": 839, "841": 840, "842": 841, "843": 842, "844": 843, "845": 844, "846": 845, "847": 846, "848": 847, "849": 848, "850": 849, "851": 850, "852": 851, "853": 852, "854": 853, "855": 854, "856": 855, "857": 856, "858": 857, "859": 858, "860": 859, "861": 860, "862": 861, "863": 862, "864": 863, "865": 864, "866": 865, "867": 866, "868": 867, "869": 868, "870": 869, "871": 870, "872": 871, "873": 872, "874": 873, "875": 874, "876": 875, "877": 876, "878": 877, "879": 878, "880": 879, "881": 880, "882": 881, "883": 882, "884": 883, "885": 884, "886": 885, "887": 886, "888": 887, "889": 888, "890": 889, "891": 890, "892": 891, "893": 892, "894": 893, "895": 894, "896": 895, "897": 896, "898": 897, "899": 898, "900": 899, "901": 900, "902": 901, "903": 902, "904": 903, "905": 904, "906": 905, "907": 906, "908": 907, "909": 908, "910": 909, "911": 910, "912": 911, "913": 912, "914": 913, "915": 914, "916": 915, "917": 916, "918": 917, "919": 918, "920": 919, "921": 920, "922": 921, "923": 922, "924": 923, "925": 924, "926": 925, "927": 926, "928": 927, "929": 928, "930": 929, "931": 930, "932": 931, "933": 932, "934": 933, "935": 934, "936": 935, "937": 936, "938": 937, "939": 938, "940": 939, "941": 940, "942": 941, "943": 942, "944": 943, "945": 944, "946": 945, "947": 946, "948": 947, "949": 948, "950": 949, "951": 950, "952": 951, "953": 952, "954": 953, "955": 954, "956": 955, "957": 956, "958": 957, "959": 958, "960": 959, "961": 960, "962": 961, "963": 962, "964": 963, "965": 964, "966": 965, "967": 966, "968": 967, "969": 968, "970": 969, "971": 970, "972": 971, "973": 972, "974": 973, "975": 974, "976": 975, "977": 976, "978": 977, "979": 978, "980": 979, "981": 980, "982": 981, "983": 982, "984": 983, "985": 984, "986": 985, "987": 986, "988": 987, "989": 988, "990": 989, "991": 990, "992": 991, "993": 992, "994": 993, "995": 994, "996": 995, "997": 996, "998": 997, "999": 998, "1000": 999, "1001": 1000, "1002": 1001, "1003": 1002, "1004": 1003, "1005": 1004, "1006": 1005, "1007": 1006, "1008": 1007, "1009": 1008, "1010": 1009, "1011": 1010, "1012": 1011, "1013": 1012, "1014": 1013, "1015": 1014, "1016": 1015, "1017": 1016, "1018": 1017, "1019": 1018, "1020": 1019, "1021": 1020, "1022": 1021, "1023": 1022, "1024": 1023, "1025": 1024, "0": null}, "code_position_map": {"1": 0, "2": 1, "3": 2, "4": 3, "5": 4, "6": 5, "7": 6, "8": 7, "9": 8, "10": 9, "11": 10, "12": 11, "13": 12, "14": 13, "15": 14, "16": 15, "17": 16, "18": 17, "19": 18, "20": 19, "21": 20, "22": 21, "23": 22, "24": 23, "0": null}, "code_pitch_map": {"1": 0, "2": 1, "3": 2, "4": 3, "5": 4, "6": 5, "7": 6, "8": 7, "9": 8, "10": 9, "11": 10, "12": 11, "13": 12, "14": 13, "15": 14, "16": 15, "17": 16, "18": 17, "19": 18, "20": 19, "21": 20, "22": 21, "23": 22, "24": 23, "25": 24, "26": 25, "27": 26, "28": 27, "29": 28, "30": 29, "31": 30, "32": 31, "33": 32, "34": 33, "35": 34, "36": 35, "37": 36, "38": 37, "39": 38, "40": 39, "41": 40, "42": 41, "43": 42, "44": 43, "45": 44, "46": 45, "47": 46, "48": 47, "49": 48, "50": 49, "51": 50, "52": 51, "53": 52, "54": 53, "55": 54, "56": 55, "57": 56, "58": 57, "59": 58, "60": 59, "61": 60, "62": 61, "63": 62, "64": 63, "65": 64, "66": 65, "67": 66, "68": 67, "69": 68, "70": 69, "71": 70, "72": 71, "73": 72, "74": 73, "75": 74, "76": 75, "77": 76, "78": 77, "79": 78, "80": 79, "81": 80, "82": 81, "83": 82, "84": 83, "85": 84, "86": 85, "87": 86, "88": 87, "89": 88, "90": 89, "91": 90, "92": 91, "93": 92, "94": 93, "95": 94, "96": 95, "97": 96, "98": 97, "99": 98, "100": 99, "101": 100, "102": 101, "103": 102, "104": 103, "105": 104, "106": 105, "107": 106, "108": 107, "109": 108, "110": 109, "111": 110, "112": 111, "113": 112, "114": 113, "115": 114, "116": 115, "117": 116, "118": 117, "119": 118, "120": 119, "121": 120, "122": 121, "123": 122, "124": 123, "125": 124, "126": 125, "127": 126, "128": 127, "0": null}, "code_duration_map": {"1": 1, "2": 2, "3": 3, "4": 12, "5": 15, "6": 16, "7": 18, "8": 20, "9": 21, "10": 24, "11": 30, "12": 36, "13": 40, "14": 42, "15": 48, "16": 60, "17": 72, "18": 84, "19": 96, "20": 120, "21": 144, "22": 168, "23": 192, "24": 384}, "code_instrument_map": {"1": "piano", "2": "electric-piano", "3": "harpsichord", "4": "clavinet", "5": "celesta", "6": "glockenspiel", "7": "music-box", "8": "vibraphone", "9": "marimba", "10": "xylophone", "11": "tubular-bells", "12": "dulcimer", "13": "organ", "14": "church-organ", "15": "accordion", "16": "harmonica", "17": "bandoneon", "18": "nylon-string-guitar", "19": "steel-string-guitar", "20": "electric-guitar", "21": "bass", "22": "electric-bass", "23": "slap-bass", "24": "synth-bass", "25": "violin", "26": "viola", "27": "cello", "28": "contrabass", "29": "harp", "30": "timpani", "31": "strings", "32": "synth-strings", "33": "voices", "34": "orchestra-hit", "35": "trumpet", "36": "trombone", "37": "tuba", "38": "horn", "39": "brasses", "40": "synth-brasses", "41": "soprano-saxophone", "42": "alto-saxophone", "43": "tenor-saxophone", "44": "baritone-saxophone", "45": "oboe", "46": "english-horn", "47": "bassoon", "48": "clarinet", "49": "piccolo", "50": "flute", "51": "recorder", "52": "pan-flute", "53": "ocarina", "54": "lead", "55": "pad", "56": "sitar", "57": "banjo", "58": "shamisen", "59": "koto", "60": "kalimba", "61": "bag-pipe", "62": "shehnai", "63": "melodic-tom", "64": "synth-drums", "0": null}, "program_instrument_map": {"0": "piano", "1": "piano", "2": "piano", "3": "piano", "4": "electric-piano", "5": "electric-piano", "6": "harpsichord", "7": "clavinet", "8": "celesta", "9": "glockenspiel", "10": "music-box", "11": "vibraphone", "12": "marimba", "13": "xylophone", "14": "tubular-bells", "15": "dulcimer", "16": "organ", "17": "organ", "18": "organ", "19": "church-organ", "20": "organ", "21": "accordion", "22": "harmonica", "23": "bandoneon", "24": "nylon-string-guitar", "25": "steel-string-guitar", "26": "electric-guitar", "27": "electric-guitar", "28": "electric-guitar", "29": "electric-guitar", "30": "electric-guitar", "31": "electric-guitar", "32": "bass", "33": "electric-bass", "34": "electric-bass", "35": "electric-bass", "36": "slap-bass", "37": "slap-bass", "38": "synth-bass", "39": "synth-bass", "40": "violin", "41": "viola", "42": "cello", "43": "contrabass", "44": "strings", "45": "strings", "46": "harp", "47": "timpani", "48": "strings", "49": "strings", "50": "synth-strings", "51": "synth-strings", "52": "voices", "53": "voices", "54": "voices", "55": "orchestra-hit", "56": "trumpet", "57": "trombone", "58": "tuba", "59": "trumpet", "60": "horn", "61": "brasses", "62": "synth-brasses", "63": "synth-brasses", "64": "soprano-saxophone", "65": "alto-saxophone", "66": "tenor-saxophone", "67": "baritone-saxophone", "68": "oboe", "69": "english-horn", "70": "bassoon", "71": "clarinet", "72": "piccolo", "73": "flute", "74": "recorder", "75": "pan-flute", "76": null, "77": null, "78": null, "79": "ocarina", "80": "lead", "81": "lead", "82": "lead", "83": "lead", "84": "lead", "85": "lead", "86": "lead", "87": "lead", "88": "pad", "89": "pad", "90": "pad", "91": "pad", "92": "pad", "93": "pad", "94": "pad", "95": "pad", "96": null, "97": null, "98": null, "99": null, "100": null, "101": null, "102": null, "103": null, "104": "sitar", "105": "banjo", "106": "shamisen", "107": "koto", "108": "kalimba", "109": "bag-pipe", "110": "violin", "111": "shehnai", "112": null, "113": null, "114": null, "115": null, "116": null, "117": "melodic-tom", "118": "synth-drums", "119": "synth-drums", "120": null, "121": null, "122": null, "123": null, "124": null, "125": null, "126": null, "127": null, "128": null}, "instrument_program_map": {"piano": 0, "electric-piano": 4, "harpsichord": 6, "clavinet": 7, "celesta": 8, "glockenspiel": 9, "music-box": 10, "vibraphone": 11, "marimba": 12, "xylophone": 13, "tubular-bells": 14, "dulcimer": 15, "organ": 16, "church-organ": 19, "accordion": 21, "harmonica": 22, "bandoneon": 23, "nylon-string-guitar": 24, "steel-string-guitar": 25, "electric-guitar": 26, "bass": 32, "electric-bass": 33, "slap-bass": 36, "synth-bass": 38, "violin": 40, "viola": 41, "cello": 42, "contrabass": 43, "harp": 46, "timpani": 47, "strings": 49, "synth-strings": 50, "voices": 52, "orchestra-hit": 55, "trumpet": 56, "trombone": 57, "tuba": 58, "horn": 60, "brasses": 61, "synth-brasses": 62, "soprano-saxophone": 64, "alto-saxophone": 65, "tenor-saxophone": 66, "baritone-saxophone": 67, "oboe": 68, "english-horn": 69, "bassoon": 70, "clarinet": 71, "piccolo": 72, "flute": 73, "recorder": 74, "pan-flute": 75, "ocarina": 79, "lead": 80, "pad": 88, "sitar": 104, "banjo": 105, "shamisen": 106, "koto": 107, "kalimba": 108, "bag-pipe": 109, "shehnai": 111, "melodic-tom": 117, "synth-drums": 118}}

mmt/music_x_transformers.py

@@ -0,0 +1,579 @@
+import argparse
+import logging
+import pathlib
+import pprint
+import sys
+
+import torch
+import torch.nn.functional as F
+from einops import repeat
+from torch import nn
+from x_transformers.autoregressive_wrapper import (
+    ENTMAX_ALPHA,
+    entmax,
+    exists,
+    top_a,
+    top_k,
+    top_p,
+)
+from x_transformers.x_transformers import (
+    AbsolutePositionalEmbedding,
+    AttentionLayers,
+    Decoder,
+    TokenEmbedding,
+    always,
+    default,
+    exists,
+)
+
+import representation
+import utils
+
+
+@utils.resolve_paths
+def parse_args(args=None, namespace=None):
+    """Parse command-line arguments."""
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-d",
+        "--dataset",
+        choices=("sod", "lmd"),
+        required=True,
+        help="dataset key",
+    )
+    parser.add_argument(
+        "-i", "--in_dir", type=pathlib.Path, help="input data directory"
+    )
+    parser.add_argument(
+        "-q", "--quiet", action="store_true", help="show warnings only"
+    )
+    return parser.parse_args(args=args, namespace=namespace)
+
+
+class MusicTransformerWrapper(nn.Module):
+    def __init__(
+        self,
+        *,
+        encoding,
+        max_seq_len,
+        attn_layers,
+        emb_dim=None,
+        max_beat=None,
+        max_mem_len=0.0,
+        shift_mem_down=0,
+        emb_dropout=0.0,
+        num_memory_tokens=None,
+        tie_embedding=False,
+        use_abs_pos_emb=True,
+        l2norm_embed=False,
+    ):
+        super().__init__()
+        assert isinstance(
+            attn_layers, AttentionLayers
+        ), "attention layers must be one of Encoder or Decoder"
+
+        dim = attn_layers.dim
+        emb_dim = default(emb_dim, dim)
+
+        self.max_seq_len = max_seq_len
+        self.max_mem_len = max_mem_len
+        self.shift_mem_down = shift_mem_down
+
+        n_tokens = encoding["n_tokens"]
+        if max_beat is not None:
+            beat_dim = encoding["dimensions"].index("beat")
+            n_tokens[beat_dim] = max_beat + 1
+
+        self.l2norm_embed = l2norm_embed
+        self.token_emb = nn.ModuleList(
+            [
+                TokenEmbedding(emb_dim, n, l2norm_embed=l2norm_embed)
+                for n in n_tokens
+            ]
+        )
+        self.pos_emb = (
+            AbsolutePositionalEmbedding(
+                emb_dim, max_seq_len, l2norm_embed=l2norm_embed
+            )
+            if (use_abs_pos_emb and not attn_layers.has_pos_emb)
+            else always(0)
+        )
+
+        self.emb_dropout = nn.Dropout(emb_dropout)
+
+        self.project_emb = (
+            nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity()
+        )
+        self.attn_layers = attn_layers
+        self.norm = nn.LayerNorm(dim)
+
+        self.init_()
+
+        self.to_logits = (
+            nn.ModuleList([nn.Linear(dim, n) for n in n_tokens])
+            if not tie_embedding
+            else [lambda t: t @ emb.weight.t() for emb in self.token_emb]
+        )
+
+        # memory tokens (like [cls]) from Memory Transformers paper
+        num_memory_tokens = default(num_memory_tokens, 0)
+        self.num_memory_tokens = num_memory_tokens
+        if num_memory_tokens > 0:
+            self.memory_tokens = nn.Parameter(
+                torch.randn(num_memory_tokens, dim)
+            )
+
+    def init_(self):
+        if self.l2norm_embed:
+            for emb in self.token_emb:
+                nn.init.normal_(emb.emb.weight, std=1e-5)
+            nn.init.normal_(self.pos_emb.emb.weight, std=1e-5)
+            return
+
+        for emb in self.token_emb:
+            nn.init.kaiming_normal_(emb.emb.weight)
+
+    def forward(
+        self,
+        x,  # shape : (b, n , d)
+        return_embeddings=False,
+        mask=None,
+        return_mems=False,
+        return_attn=False,
+        mems=None,
+        **kwargs,
+    ):
+        b, _, _ = x.shape
+        num_mem = self.num_memory_tokens
+
+        x = sum(
+            emb(x[..., i]) for i, emb in enumerate(self.token_emb)
+        ) + self.pos_emb(x)
+        x = self.emb_dropout(x)
+
+        x = self.project_emb(x)
+
+        if num_mem > 0:
+            mem = repeat(self.memory_tokens, "n d -> b n d", b=b)
+            x = torch.cat((mem, x), dim=1)
+
+            # auto-handle masking after appending memory tokens
+            if exists(mask):
+                mask = F.pad(mask, (num_mem, 0), value=True)
+
+        if self.shift_mem_down and exists(mems):
+            mems_l, mems_r = (
+                mems[: self.shift_mem_down],
+                mems[self.shift_mem_down :],
+            )
+            mems = [*mems_r, *mems_l]
+
+        x, intermediates = self.attn_layers(
+            x, mask=mask, mems=mems, return_hiddens=True, **kwargs
+        )
+        x = self.norm(x)
+
+        mem, x = x[:, :num_mem], x[:, num_mem:]
+
+        out = (
+            [to_logit(x) for to_logit in self.to_logits]
+            if not return_embeddings
+            else x
+        )
+
+        if return_mems:
+            hiddens = intermediates.hiddens
+            new_mems = (
+                list(
+                    map(
+                        lambda pair: torch.cat(pair, dim=-2),
+                        zip(mems, hiddens),
+                    )
+                )
+                if exists(mems)
+                else hiddens
+            )
+            new_mems = list(
+                map(
+                    lambda t: t[..., -self.max_mem_len :, :].detach(), new_mems
+                )
+            )
+            return out, new_mems
+
+        if return_attn:
+            attn_maps = list(
+                map(
+                    lambda t: t.post_softmax_attn,
+                    intermediates.attn_intermediates,
+                )
+            )
+            return out, attn_maps
+
+        return out
+
+
+def sample(logits, kind, threshold, temperature, min_p_pow, min_p_ratio):
+    """Sample from the logits with a specific sampling strategy."""
+    if kind == "top_k":
+        probs = F.softmax(top_k(logits, thres=threshold) / temperature, dim=-1)
+    elif kind == "top_p":
+        probs = F.softmax(top_p(logits, thres=threshold) / temperature, dim=-1)
+    elif kind == "top_a":
+        probs = F.softmax(
+            top_a(logits, min_p_pow=min_p_pow, min_p_ratio=min_p_ratio)
+            / temperature,
+            dim=-1,
+        )
+    elif kind == "entmax":
+        probs = entmax(logits / temperature, alpha=ENTMAX_ALPHA, dim=-1)
+    else:
+        raise ValueError(f"Unknown sampling strategy: {kind}")
+
+    return torch.multinomial(probs, 1)
+
+
+class MusicAutoregressiveWrapper(nn.Module):
+    def __init__(self, net, encoding, ignore_index=-100, pad_value=0):
+        super().__init__()
+        self.pad_value = pad_value
+        self.ignore_index = ignore_index
+
+        self.net = net
+        self.max_seq_len = net.max_seq_len
+
+        # Get the type codes
+        self.sos_type_code = encoding["type_code_map"]["start-of-song"]
+        self.eos_type_code = encoding["type_code_map"]["end-of-song"]
+        self.son_type_code = encoding["type_code_map"]["start-of-notes"]
+        self.instrument_type_code = encoding["type_code_map"]["instrument"]
+        self.note_type_code = encoding["type_code_map"]["note"]
+
+        # Get the dimension indices
+        self.dimensions = {
+            key: encoding["dimensions"].index(key)
+            for key in (
+                "type",
+                "beat",
+                "position",
+                "pitch",
+                "duration",
+                "instrument",
+            )
+        }
+        assert self.dimensions["type"] == 0
+
+    @torch.no_grad()
+    def generate(
+        self,
+        start_tokens,  # shape : (b, n, d)
+        seq_len,
+        eos_token=None,
+        temperature=1.0,  # int or list of int
+        filter_logits_fn="top_k",  # str or list of str
+        filter_thres=0.9,  # int or list of int
+        min_p_pow=2.0,
+        min_p_ratio=0.02,
+        monotonicity_dim=None,
+        return_attn=False,
+        **kwargs,
+    ):
+        _, t, dim = start_tokens.shape
+
+        if isinstance(temperature, (float, int)):
+            temperature = [temperature] * dim
+        elif len(temperature) == 1:
+            temperature = temperature * dim
+        else:
+            assert (
+                len(temperature) == dim
+            ), f"`temperature` must be of length {dim}"
+
+        if isinstance(filter_logits_fn, str):
+            filter_logits_fn = [filter_logits_fn] * dim
+        elif len(filter_logits_fn) == 1:
+            filter_logits_fn = filter_logits_fn * dim
+        else:
+            assert (
+                len(filter_logits_fn) == dim
+            ), f"`filter_logits_fn` must be of length {dim}"
+
+        if isinstance(filter_thres, (float, int)):
+            filter_thres = [filter_thres] * dim
+        elif len(filter_thres) == 1:
+            filter_thres = filter_thres * dim
+        else:
+            assert (
+                len(filter_thres) == dim
+            ), f"`filter_thres` must be of length {dim}"
+
+        if isinstance(monotonicity_dim, str):
+            monotonicity_dim = [self.dimensions[monotonicity_dim]]
+        else:
+            monotonicity_dim = [self.dimensions[d] for d in monotonicity_dim]
+
+        was_training = self.net.training
+        num_dims = len(start_tokens.shape)
+
+        if num_dims == 2:
+            start_tokens = start_tokens[None, :, :]
+
+        self.net.eval()
+        out = start_tokens
+        mask = kwargs.pop("mask", None)
+
+        if mask is None:
+            mask = torch.ones(
+                (out.shape[0], out.shape[1]),
+                dtype=torch.bool,
+                device=out.device,
+            )
+
+        if monotonicity_dim is not None:
+            current_values = {
+                d: torch.max(start_tokens[:, :, d], 1)[0]
+                for d in monotonicity_dim
+            }
+        else:
+            current_values = None
+
+        instrument_dim = self.dimensions["instrument"]
+        type_dim = self.dimensions["type"]
+        for _ in range(seq_len):
+            x = out[:, -self.max_seq_len :]
+            mask = mask[:, -self.max_seq_len :]
+
+            if return_attn:
+                logits, attn = self.net(
+                    x, mask=mask, return_attn=True, **kwargs
+                )
+                logits = [l[:, -1, :] for l in logits]
+            else:
+                logits = [
+                    l[:, -1, :] for l in self.net(x, mask=mask, **kwargs)
+                ]
+
+            # Enforce monotonicity
+            if monotonicity_dim is not None and 0 in monotonicity_dim:
+                for i, v in enumerate(current_values[0]):
+                    logits[0][i, :v] = -float("inf")
+
+            # Filter out sos token
+            logits[0][type_dim, 0] = -float("inf")
+
+            # Sample from the logits
+            sample_type = sample(
+                logits[0],
+                filter_logits_fn[0],
+                filter_thres[0],
+                temperature[0],
+                min_p_pow,
+                min_p_ratio,
+            )
+
+            # Update current values
+            if monotonicity_dim is not None and 0 in monotonicity_dim:
+                current_values[0] = torch.maximum(
+                    current_values[0], sample_type.reshape(-1)
+                )
+
+            # Iterate after each sample
+            samples = [[s_type] for s_type in sample_type]
+            for idx, s_type in enumerate(sample_type):
+                # A start-of-song, end-of-song or start-of-notes code
+                if s_type in (
+                    self.sos_type_code,
+                    self.eos_type_code,
+                    self.son_type_code,
+                ):
+                    samples[idx] += [torch.zeros_like(s_type)] * (
+                        len(logits) - 1
+                    )
+                # An instrument code
+                elif s_type == self.instrument_type_code:
+                    samples[idx] += [torch.zeros_like(s_type)] * (
+                        len(logits) - 2
+                    )
+                    logits[instrument_dim][:, 0] = -float("inf")  # avoid none
+                    sampled = sample(
+                        logits[instrument_dim][idx : idx + 1],
+                        filter_logits_fn[instrument_dim],
+                        filter_thres[instrument_dim],
+                        temperature[instrument_dim],
+                        min_p_pow,
+                        min_p_ratio,
+                    )[0]
+                    samples[idx].append(sampled)
+                # A note code
+                elif s_type == self.note_type_code:
+                    for d in range(1, dim):
+                        # Enforce monotonicity
+                        if (
+                            monotonicity_dim is not None
+                            and d in monotonicity_dim
+                        ):
+                            logits[d][idx, : current_values[d][idx]] = -float(
+                                "inf"
+                            )
+
+                        # Sample from the logits
+                        logits[d][:, 0] = -float("inf")  # avoid none
+                        sampled = sample(
+                            logits[d][idx : idx + 1],
+                            filter_logits_fn[d],
+                            filter_thres[d],
+                            temperature[d],
+                            min_p_pow,
+                            min_p_ratio,
+                        )[0]
+                        samples[idx].append(sampled)
+
+                        # Update current values
+                        if (
+                            monotonicity_dim is not None
+                            and d in monotonicity_dim
+                        ):
+                            current_values[d][idx] = torch.max(
+                                current_values[d][idx], sampled
+                            )[0]
+                else:
+                    raise ValueError(f"Unknown event type code: {s_type}")
+
+            stacked = torch.stack(
+                [torch.cat(s).expand(1, -1) for s in samples], 0
+            )
+            out = torch.cat((out, stacked), dim=1)
+            mask = F.pad(mask, (0, 1), value=True)
+
+            if exists(eos_token):
+                is_eos_tokens = out[..., 0] == eos_token
+
+                # Mask out everything after the eos tokens
+                if is_eos_tokens.any(dim=1).all():
+                    for i, is_eos_token in enumerate(is_eos_tokens):
+                        idx = torch.argmax(is_eos_token.byte())
+                        out[i, idx + 1 :] = self.pad_value
+                    break
+
+        out = out[:, t:]
+
+        if num_dims == 1:
+            out = out.squeeze(0)
+
+        self.net.train(was_training)
+
+        if return_attn:
+            return out, attn
+
+        return out
+
+    def forward(self, x, return_list=False, **kwargs):
+        xi = x[:, :-1]
+        xo = x[:, 1:]
+
+        # help auto-solve a frequent area of confusion around input masks in auto-regressive
+        # if user supplies a mask that is only off by one from the source sequence, resolve it for them
+        mask = kwargs.get("mask", None)
+        if mask is not None and mask.shape[1] == x.shape[1]:
+            mask = mask[:, :-1]
+            kwargs["mask"] = mask
+
+        out = self.net(xi, **kwargs)
+        losses = [
+            F.cross_entropy(
+                out[i].transpose(1, 2),
+                xo[..., i],
+                ignore_index=self.ignore_index,
+            )
+            for i in range(len(out))
+        ]
+        loss = sum(losses)
+
+        if return_list:
+            return loss, losses
+        return loss
+
+
+class MusicXTransformer(nn.Module):
+    def __init__(self, *, dim, encoding, **kwargs):
+        super().__init__()
+        assert "dim" not in kwargs, "dimension must be set with `dim` keyword"
+        transformer_kwargs = {
+            "max_seq_len": kwargs.pop("max_seq_len"),
+            "max_beat": kwargs.pop("max_beat"),
+            "emb_dropout": kwargs.pop("emb_dropout", 0),
+            "use_abs_pos_emb": kwargs.pop("use_abs_pos_emb", True),
+        }
+        self.decoder = MusicTransformerWrapper(
+            encoding=encoding,
+            attn_layers=Decoder(dim=dim, **kwargs),
+            **transformer_kwargs,
+        )
+        self.decoder = MusicAutoregressiveWrapper(
+            self.decoder, encoding=encoding
+        )
+
+    @torch.no_grad()
+    def generate(self, seq_in, seq_len, **kwargs):
+        return self.decoder.generate(seq_in, seq_len, **kwargs)
+
+    def forward(self, seq, mask=None, **kwargs):
+        return self.decoder(seq, mask=mask, **kwargs)
+
+
+def main():
+    """Main function."""
+    # Parse the command-line arguments
+    args = parse_args()
+
+    # Set default arguments
+    if args.dataset is not None:
+        if args.in_dir is None:
+            args.in_dir = pathlib.Path(f"data/{args.dataset}/processed/notes")
+
+    # Set up the logger
+    logging.basicConfig(
+        stream=sys.stdout,
+        level=logging.ERROR if args.quiet else logging.INFO,
+        format="%(levelname)-8s %(message)s",
+    )
+
+    # Log arguments
+    logging.info(f"Using arguments:\n{pprint.pformat(vars(args))}")
+
+    # Load the encoding
+    encoding = representation.load_encoding(args.in_dir / "encoding.json")
+
+    # Create the model
+    model = MusicXTransformer(
+        dim=128,
+        encoding=encoding,
+        depth=3,
+        heads=4,
+        max_seq_len=1024,
+        max_beat=256,
+        rel_pos_bias=True,  # relative positional bias
+        rotary_pos_emb=True,  # rotary positional encoding
+        emb_dropout=0.1,
+        attn_dropout=0.1,
+        ff_dropout=0.1,
+    )
+
+    # Summarize the model
+    n_parameters = sum(p.numel() for p in model.parameters())
+    n_trainables = sum(
+        p.numel() for p in model.parameters() if p.requires_grad
+    )
+    print(f"Number of parameters: {n_parameters}")
+    print(f"Number of trainable parameters: {n_trainables}")
+
+    # Create test data
+    seq = torch.randint(0, 4, (1, 1024, 6))
+    mask = torch.ones((1, 1024)).bool()
+
+    # Pass test data through the model
+    loss = model(seq, mask=mask)
+    loss.backward()
+
+
+if __name__ == "__main__":
+    main()

mmt/representation.py

@@ -0,0 +1,756 @@
+"""Representation utilities."""
+import pathlib
+import pprint
+
+import muspy
+import numpy as np
+import pretty_midi
+
+import utils
+
+# Configuration
+RESOLUTION = 12
+MAX_BEAT = 1024
+MAX_DURATION = 384  # Remember to modify known durations as well!
+
+# Dimensions
+# (NOTE: "type" must be the first dimension!)
+# (NOTE: Remember to modify N_TOKENS as well!)
+DIMENSIONS = ["type", "beat", "position", "pitch", "duration", "instrument"]
+assert DIMENSIONS[0] == "type"
+
+# Type
+TYPE_CODE_MAP = {
+    "start-of-song": 0,
+    "instrument": 1,
+    "start-of-notes": 2,
+    "note": 3,
+    "end-of-song": 4,
+}
+CODE_TYPE_MAP = utils.inverse_dict(TYPE_CODE_MAP)
+
+# Beat
+BEAT_CODE_MAP = {i: i + 1 for i in range(MAX_BEAT + 1)}
+BEAT_CODE_MAP[None] = 0
+CODE_BEAT_MAP = utils.inverse_dict(BEAT_CODE_MAP)
+
+# Position
+POSITION_CODE_MAP = {i: i + 1 for i in range(RESOLUTION)}
+POSITION_CODE_MAP[None] = 0
+CODE_POSITION_MAP = utils.inverse_dict(POSITION_CODE_MAP)
+
+# Pitch
+PITCH_CODE_MAP = {i: i + 1 for i in range(128)}
+PITCH_CODE_MAP[None] = 0
+CODE_PITCH_MAP = utils.inverse_dict(PITCH_CODE_MAP)
+
+# Duration
+KNOWN_DURATIONS = [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    9,
+    10,
+    11,
+    12,
+    15,
+    16,
+    18,
+    20,
+    21,
+    24,
+    30,
+    36,
+    40,
+    42,
+    48,
+    60,
+    72,
+    84,
+    96,
+    120,
+    144,
+    168,
+    192,
+    384,
+]
+DURATION_CODE_MAP = {
+    i: int(np.argmin(np.abs(np.array(KNOWN_DURATIONS) - i))) + 1
+    for i in range(MAX_DURATION + 1)
+}
+DURATION_CODE_MAP[None] = 0
+CODE_DURATION_MAP = {
+    i + 1: duration for i, duration in enumerate(KNOWN_DURATIONS)
+}
+
+# Instrument
+PROGRAM_INSTRUMENT_MAP = {
+    # Pianos
+    0: "piano",
+    1: "piano",
+    2: "piano",
+    3: "piano",
+    4: "electric-piano",
+    5: "electric-piano",
+    6: "harpsichord",
+    7: "clavinet",
+    # Chromatic Percussion
+    8: "celesta",
+    9: "glockenspiel",
+    10: "music-box",
+    11: "vibraphone",
+    12: "marimba",
+    13: "xylophone",
+    14: "tubular-bells",
+    15: "dulcimer",
+    # Organs
+    16: "organ",
+    17: "organ",
+    18: "organ",
+    19: "church-organ",
+    20: "organ",
+    21: "accordion",
+    22: "harmonica",
+    23: "bandoneon",
+    # Guitars
+    24: "nylon-string-guitar",
+    25: "steel-string-guitar",
+    26: "electric-guitar",
+    27: "electric-guitar",
+    28: "electric-guitar",
+    29: "electric-guitar",
+    30: "electric-guitar",
+    31: "electric-guitar",
+    # Basses
+    32: "bass",
+    33: "electric-bass",
+    34: "electric-bass",
+    35: "electric-bass",
+    36: "slap-bass",
+    37: "slap-bass",
+    38: "synth-bass",
+    39: "synth-bass",
+    # Strings
+    40: "violin",
+    41: "viola",
+    42: "cello",
+    43: "contrabass",
+    44: "strings",
+    45: "strings",
+    46: "harp",
+    47: "timpani",
+    # Ensemble
+    48: "strings",
+    49: "strings",
+    50: "synth-strings",
+    51: "synth-strings",
+    52: "voices",
+    53: "voices",
+    54: "voices",
+    55: "orchestra-hit",
+    # Brass
+    56: "trumpet",
+    57: "trombone",
+    58: "tuba",
+    59: "trumpet",
+    60: "horn",
+    61: "brasses",
+    62: "synth-brasses",
+    63: "synth-brasses",
+    # Reed
+    64: "soprano-saxophone",
+    65: "alto-saxophone",
+    66: "tenor-saxophone",
+    67: "baritone-saxophone",
+    68: "oboe",
+    69: "english-horn",
+    70: "bassoon",
+    71: "clarinet",
+    # Pipe
+    72: "piccolo",
+    73: "flute",
+    74: "recorder",
+    75: "pan-flute",
+    76: None,
+    77: None,
+    78: None,
+    79: "ocarina",
+    # Synth Lead
+    80: "lead",
+    81: "lead",
+    82: "lead",
+    83: "lead",
+    84: "lead",
+    85: "lead",
+    86: "lead",
+    87: "lead",
+    # Synth Pad
+    88: "pad",
+    89: "pad",
+    90: "pad",
+    91: "pad",
+    92: "pad",
+    93: "pad",
+    94: "pad",
+    95: "pad",
+    # Synth Effects
+    96: None,
+    97: None,
+    98: None,
+    99: None,
+    100: None,
+    101: None,
+    102: None,
+    103: None,
+    # Ethnic
+    104: "sitar",
+    105: "banjo",
+    106: "shamisen",
+    107: "koto",
+    108: "kalimba",
+    109: "bag-pipe",
+    110: "violin",
+    111: "shehnai",
+    # Percussive
+    112: None,
+    113: None,
+    114: None,
+    115: None,
+    116: None,
+    117: "melodic-tom",
+    118: "synth-drums",
+    119: "synth-drums",
+    # Sound effects
+    120: None,
+    121: None,
+    122: None,
+    123: None,
+    124: None,
+    125: None,
+    126: None,
+    127: None,
+}
+INSTRUMENT_PROGRAM_MAP = {
+    # Pianos
+    "piano": 0,
+    "electric-piano": 4,
+    "harpsichord": 6,
+    "clavinet": 7,
+    # Chromatic Percussion
+    "celesta": 8,
+    "glockenspiel": 9,
+    "music-box": 10,
+    "vibraphone": 11,
+    "marimba": 12,
+    "xylophone": 13,
+    "tubular-bells": 14,
+    "dulcimer": 15,
+    # Organs
+    "organ": 16,
+    "church-organ": 19,
+    "accordion": 21,
+    "harmonica": 22,
+    "bandoneon": 23,
+    # Guitars
+    "nylon-string-guitar": 24,
+    "steel-string-guitar": 25,
+    "electric-guitar": 26,
+    # Basses
+    "bass": 32,
+    "electric-bass": 33,
+    "slap-bass": 36,
+    "synth-bass": 38,
+    # Strings
+    "violin": 40,
+    "viola": 41,
+    "cello": 42,
+    "contrabass": 43,
+    "harp": 46,
+    "timpani": 47,
+    # Ensemble
+    "strings": 49,
+    "synth-strings": 50,
+    "voices": 52,
+    "orchestra-hit": 55,
+    # Brass
+    "trumpet": 56,
+    "trombone": 57,
+    "tuba": 58,
+    "horn": 60,
+    "brasses": 61,
+    "synth-brasses": 62,
+    # Reed
+    "soprano-saxophone": 64,
+    "alto-saxophone": 65,
+    "tenor-saxophone": 66,
+    "baritone-saxophone": 67,
+    "oboe": 68,
+    "english-horn": 69,
+    "bassoon": 70,
+    "clarinet": 71,
+    # Pipe
+    "piccolo": 72,
+    "flute": 73,
+    "recorder": 74,
+    "pan-flute": 75,
+    "ocarina": 79,
+    # Synth Lead
+    "lead": 80,
+    # Synth Pad
+    "pad": 88,
+    # Ethnic
+    "sitar": 104,
+    "banjo": 105,
+    "shamisen": 106,
+    "koto": 107,
+    "kalimba": 108,
+    "bag-pipe": 109,
+    "shehnai": 111,
+    # Percussive
+    "melodic-tom": 117,
+    "synth-drums": 118,
+}
+KNOWN_PROGRAMS = list(
+    k for k, v in PROGRAM_INSTRUMENT_MAP.items() if v is not None
+)
+KNOWN_INSTRUMENTS = list(dict.fromkeys(INSTRUMENT_PROGRAM_MAP.keys()))
+INSTRUMENT_CODE_MAP = {
+    instrument: i + 1 for i, instrument in enumerate(KNOWN_INSTRUMENTS)
+}
+INSTRUMENT_CODE_MAP[None] = 0
+CODE_INSTRUMENT_MAP = utils.inverse_dict(INSTRUMENT_CODE_MAP)
+
+N_TOKENS = [
+    max(TYPE_CODE_MAP.values()) + 1,
+    max(BEAT_CODE_MAP.values()) + 1,
+    max(POSITION_CODE_MAP.values()) + 1,
+    max(PITCH_CODE_MAP.values()) + 1,
+    max(DURATION_CODE_MAP.values()) + 1,
+    max(INSTRUMENT_CODE_MAP.values()) + 1,
+]
+
+
+def get_encoding():
+    """Return the encoding configurations."""
+    return {
+        "resolution": RESOLUTION,
+        "max_beat": MAX_BEAT,
+        "max_duration": MAX_DURATION,
+        "dimensions": DIMENSIONS,
+        "n_tokens": N_TOKENS,
+        "type_code_map": TYPE_CODE_MAP,
+        "beat_code_map": BEAT_CODE_MAP,
+        "position_code_map": POSITION_CODE_MAP,
+        "pitch_code_map": PITCH_CODE_MAP,
+        "duration_code_map": DURATION_CODE_MAP,
+        "instrument_code_map": INSTRUMENT_CODE_MAP,
+        "code_type_map": CODE_TYPE_MAP,
+        "code_beat_map": CODE_BEAT_MAP,
+        "code_position_map": CODE_POSITION_MAP,
+        "code_pitch_map": CODE_PITCH_MAP,
+        "code_duration_map": CODE_DURATION_MAP,
+        "code_instrument_map": CODE_INSTRUMENT_MAP,
+        "program_instrument_map": PROGRAM_INSTRUMENT_MAP,
+        "instrument_program_map": INSTRUMENT_PROGRAM_MAP,
+    }
+
+
+def load_encoding(filename):
+    """Load encoding configurations from a JSON file."""
+    encoding = utils.load_json(filename)
+    for key in (
+        "code_type_map",
+        "code_beat_map",
+        "code_position_map",
+        "code_duration_map",
+        "code_pitch_map",
+        "code_instrument_map",
+        "beat_code_map",
+        "position_code_map",
+        "duration_code_map",
+        "pitch_code_map",
+        "program_instrument_map",
+    ):
+        encoding[key] = {
+            int(k) if k != "null" else None: v
+            for k, v in encoding[key].items()
+        }
+    return encoding
+
+
+def extract_notes(music, resolution):
+    """Return a MusPy music object as a note sequence.
+
+    Each row of the output is a note specified as follows.
+
+        (beat, position, pitch, duration, program)
+
+    """
+    # Check resolution
+    assert music.resolution == resolution
+
+    # Extract notes
+    notes = []
+    for track in music:
+        if track.is_drum or track.program not in KNOWN_PROGRAMS:
+            continue
+        for note in track:
+            beat, position = divmod(note.time, resolution)
+            notes.append(
+                (beat, position, note.pitch, note.duration, track.program)
+            )
+
+    # Deduplicate and sort the notes
+    notes = sorted(set(notes))
+
+    return np.array(notes)
+
+
+def encode_notes(notes, encoding):
+    """Encode a note sequence into a sequence of codes.
+
+    Each row of the input is a note specified as follows.
+
+        (beat, position, pitch, duration, program)
+
+    Each row of the output is encoded as follows.
+
+        (event_type, beat, position, pitch, duration, instrument)
+
+    """
+    # Get variables
+    max_beat = encoding["max_beat"]
+    max_duration = encoding["max_duration"]
+
+    # Get maps
+    type_code_map = encoding["type_code_map"]
+    beat_code_map = encoding["beat_code_map"]
+    position_code_map = encoding["position_code_map"]
+    pitch_code_map = encoding["pitch_code_map"]
+    duration_code_map = encoding["duration_code_map"]
+    instrument_code_map = encoding["instrument_code_map"]
+    program_instrument_map = encoding["program_instrument_map"]
+
+    # Get the dimension indices
+    beat_dim = encoding["dimensions"].index("beat")
+    position_dim = encoding["dimensions"].index("position")
+    pitch_dim = encoding["dimensions"].index("pitch")
+    duration_dim = encoding["dimensions"].index("duration")
+    instrument_dim = encoding["dimensions"].index("instrument")
+
+    # Start the codes with an SOS row
+    codes = [(type_code_map["start-of-song"], 0, 0, 0, 0, 0)]
+
+    # Extract instruments
+    instruments = set(program_instrument_map[note[-1]] for note in notes)
+
+    # Encode the instruments
+    instrument_codes = []
+    for instrument in instruments:
+        # Skip unknown instruments
+        if instrument is None:
+            continue
+        row = [type_code_map["instrument"], 0, 0, 0, 0, 0]
+        row[instrument_dim] = instrument_code_map[instrument]
+        instrument_codes.append(row)
+
+    # Sort the instruments and append them to the code sequence
+    instrument_codes.sort()
+    codes.extend(instrument_codes)
+
+    # Encode the notes
+    codes.append((type_code_map["start-of-notes"], 0, 0, 0, 0, 0))
+    for beat, position, pitch, duration, program in notes:
+        # Skip if max_beat has reached
+        if beat > max_beat:
+            continue
+        # Skip unknown instruments
+        instrument = program_instrument_map[program]
+        if instrument is None:
+            continue
+        # Encode the note
+        row = [type_code_map["note"], 0, 0, 0, 0, 0]
+        row[beat_dim] = beat_code_map[beat]
+        row[position_dim] = position_code_map[position]
+        row[pitch_dim] = pitch_code_map[pitch]
+        row[duration_dim] = duration_code_map[min(duration, max_duration)]
+        row[instrument_dim] = instrument_code_map[instrument]
+        codes.append(row)
+
+    # End the codes with an EOS row
+    codes.append((type_code_map["end-of-song"], 0, 0, 0, 0, 0))
+
+    return np.array(codes)
+
+
+def encode(music, encoding):
+    """Encode a MusPy music object into a sequence of codes.
+
+    Each row of the input is encoded as follows.
+
+        (event_type, beat, position, pitch, duration, instrument)
+
+    Each row of the output is a note specified as follows.
+
+        (beat, position, pitch, duration, program)
+
+    """
+    # Extract notes
+    notes = extract_notes(music, encoding["resolution"])
+
+    # Encode the notes
+    codes = encode_notes(notes, encoding)
+
+    return codes
+
+
+def decode_notes(codes, encoding):
+    """Decode codes into a note sequence.
+
+    Each row of the input is encoded as follows.
+
+        (event_type, beat, position, pitch, duration, instrument)
+
+    """
+    # Get variables and maps
+    code_type_map = encoding["code_type_map"]
+    code_beat_map = encoding["code_beat_map"]
+    code_position_map = encoding["code_position_map"]
+    code_pitch_map = encoding["code_pitch_map"]
+    code_duration_map = encoding["code_duration_map"]
+    code_instrument_map = encoding["code_instrument_map"]
+    instrument_program_map = encoding["instrument_program_map"]
+
+    # Get the dimension indices
+    beat_dim = encoding["dimensions"].index("beat")
+    position_dim = encoding["dimensions"].index("position")
+    pitch_dim = encoding["dimensions"].index("pitch")
+    duration_dim = encoding["dimensions"].index("duration")
+    instrument_dim = encoding["dimensions"].index("instrument")
+
+    # Decode the codes into a sequence of notes
+    notes = []
+    for row in codes:
+        event_type = code_type_map[int(row[0])]
+        if event_type in ("start-of-song", "instrument", "start-of-notes"):
+            continue
+        elif event_type == "end-of-song":
+            break
+        elif event_type == "note":
+            beat = code_beat_map[int(row[beat_dim])]
+            position = code_position_map[int(row[position_dim])]
+            pitch = code_pitch_map[int(row[pitch_dim])]
+            duration = code_duration_map[int(row[duration_dim])]
+            instrument = code_instrument_map[int(row[instrument_dim])]
+            program = instrument_program_map[instrument]
+            notes.append((beat, position, pitch, duration, program))
+        else:
+            raise ValueError("Unknown event type.")
+
+    return notes
+
+
+def reconstruct(notes, resolution):
+    """Reconstruct a note sequence to a MusPy Music object."""
+    # Construct the MusPy Music object
+    music = muspy.Music(resolution=resolution, tempos=[muspy.Tempo(0, 100)])
+
+    # Append the tracks
+    programs = sorted(set(note[-1] for note in notes))
+    for program in programs:
+        music.tracks.append(muspy.Track(program))
+
+    # Append the notes
+    for beat, position, pitch, duration, program in notes:
+        time = beat * resolution + position
+        track_idx = programs.index(program)
+        music[track_idx].notes.append(muspy.Note(time, pitch, duration))
+
+    return music
+
+
+def decode(codes, encoding):
+    """Decode codes into a MusPy Music object.
+
+    Each row of the input is encoded as follows.
+
+        (event_type, beat, position, pitch, duration, instrument)
+
+    """
+    # Get resolution
+    resolution = encoding["resolution"]
+
+    # Decode codes into a note sequence
+    notes = decode_notes(codes, encoding)
+
+    # Reconstruct the music object
+    music = reconstruct(notes, resolution)
+
+    return music
+
+
+def dump(data, encoding):
+    """Decode the codes and dump as a string."""
+    # Get maps
+    code_type_map = encoding["code_type_map"]
+    code_beat_map = encoding["code_beat_map"]
+    code_position_map = encoding["code_position_map"]
+    code_pitch_map = encoding["code_pitch_map"]
+    code_duration_map = encoding["code_duration_map"]
+    code_instrument_map = encoding["code_instrument_map"]
+
+    # Get the dimension indices
+    beat_dim = encoding["dimensions"].index("beat")
+    position_dim = encoding["dimensions"].index("position")
+    pitch_dim = encoding["dimensions"].index("pitch")
+    duration_dim = encoding["dimensions"].index("duration")
+    instrument_dim = encoding["dimensions"].index("instrument")
+
+    # Iterate over the rows
+    lines = []
+    for row in data:
+        event_type = code_type_map[int(row[0])]
+        if event_type == "start-of-song":
+            lines.append("Start of song")
+        elif event_type == "end-of-song":
+            lines.append("End of song")
+        elif event_type == "instrument":
+            instrument = code_instrument_map[int(row[instrument_dim])]
+            lines.append(f"Instrument: {instrument}")
+        elif event_type == "start-of-notes":
+            lines.append("Start of notes")
+        elif event_type == "note":
+            beat = code_beat_map[int(row[beat_dim])]
+            position = code_position_map[int(row[position_dim])]
+            pitch = pretty_midi.note_number_to_name(
+                code_pitch_map[int(row[pitch_dim])]
+            )
+            duration = code_duration_map[int(row[duration_dim])]
+            instrument = code_instrument_map[int(row[instrument_dim])]
+            lines.append(
+                f"Note: beat={beat}, position={position}, pitch={pitch}, "
+                f"duration={duration}, instrument={instrument}"
+            )
+        else:
+            raise ValueError(f"Unknown event type: {event_type}")
+
+    return "\n".join(lines)
+
+
+def save_txt(filename, data, encoding):
+    """Dump the codes into a TXT file."""
+    with open(filename, "w") as f:
+        f.write(dump(data, encoding))
+
+
+def save_csv_notes(filename, data):
+    """Save the representation as a CSV file."""
+    assert data.shape[1] == 5
+    np.savetxt(
+        filename,
+        data,
+        fmt="%d",
+        delimiter=",",
+        header="beat,position,pitch,duration,program",
+        comments="",
+    )
+
+
+def save_csv_codes(filename, data):
+    """Save the representation as a CSV file."""
+    assert data.shape[1] == 6
+    np.savetxt(
+        filename,
+        data,
+        fmt="%d",
+        delimiter=",",
+        header="type,beat,position,pitch,duration,instrument",
+        comments="",
+    )
+
+
+def main():
+    """Main function."""
+    # Get the encoding
+    encoding = get_encoding()
+
+    # Save the encoding
+    filename = pathlib.Path(__file__).parent / "encoding.json"
+    utils.save_json(filename, encoding)
+
+    # Load encoding
+    encoding = load_encoding(filename)
+
+    # Print the maps
+    print(f"{' Maps ':=^40}")
+    for key, value in encoding.items():
+        if key in (
+            "instrument_code_map",
+            "code_instrument_map",
+            "program_instrument_map",
+            "instrument_program_map",
+        ):
+            print("-" * 40)
+            print(f"{key}:")
+            pprint.pprint(value, indent=2)
+
+    # Print the variables
+    print(f"{' Variables ':=^40}")
+    print(f"resolution: {encoding['resolution']}")
+    print(f"max_beat: {encoding['max_beat']}")
+    print(f"max_duration: {encoding['max_duration']}")
+
+    # Print the number of tokens
+    print(f"{' Number of tokens ':=^40}")
+    keys = ("type", "beat", "position", "pitch", "duration", "instrument")
+    for key, value in zip(keys, N_TOKENS):
+        print(f"{key}: {value}")
+
+    # Print an example
+    print(f"{'Example':=^40}")
+    codes = np.array(
+        (
+            (0, 0, 0, 0, 0, 0),
+            (1, 0, 0, 0, 0, 3),
+            (1, 0, 0, 0, 0, 33),
+            (2, 0, 0, 0, 0, 0),
+            (3, 1, 1, 49, 15, 3),
+            (3, 1, 1, 61, 15, 3),
+            (3, 1, 1, 65, 15, 3),
+            (3, 1, 1, 68, 10, 33),
+            (3, 1, 1, 68, 15, 3),
+            (3, 2, 1, 68, 10, 33),
+            (3, 3, 1, 68, 10, 33),
+            (3, 4, 1, 61, 10, 33),
+            (3, 4, 1, 61, 15, 3),
+            (3, 4, 1, 65, 4, 33),
+            (3, 4, 1, 65, 10, 3),
+            (3, 4, 1, 68, 10, 3),
+            (3, 4, 1, 73, 10, 3),
+            (3, 4, 13, 63, 4, 33),
+            (4, 0, 0, 0, 0, 0),
+        ),
+        int,
+    )
+    print(f"Codes:\n{codes}")
+
+    print("-" * 40)
+    print(f"Decoded:\n{dump(codes, encoding)}")
+
+    music = decode(codes, encoding)
+    print("-" * 40)
+    print(f"Decoded music:\n{music}")
+
+    encoded = encode(music, encoding)
+    print("-" * 40)
+    print(f"Encoded:\n{encoded}")
+    assert np.all(codes == encoded)
+
+
+if __name__ == "__main__":
+    main()

mmt/utils.py

@@ -0,0 +1,88 @@
+"""Utility functions."""
+import json
+import pathlib
+import warnings
+
+import numpy as np
+
+
+def save_args(filename, args):
+    """Save the command-line arguments."""
+    args_dict = {}
+    for key, value in vars(args).items():
+        if isinstance(value, pathlib.Path):
+            args_dict[key] = str(value)
+        else:
+            args_dict[key] = value
+    save_json(filename, args_dict)
+
+
+def inverse_dict(d):
+    """Return the inverse dictionary."""
+    return {v: k for k, v in d.items()}
+
+
+def save_txt(filename, data):
+    """Save a list to a TXT file."""
+    with open(filename, "w", encoding="utf8") as f:
+        for item in data:
+            f.write(f"{item}\n")
+
+
+def load_txt(filename):
+    """Load a TXT file as a list."""
+    with open(filename, encoding="utf8") as f:
+        return [line.strip() for line in f]
+
+
+def save_json(filename, data):
+    """Save data as a JSON file."""
+    with open(filename, "w", encoding="utf8") as f:
+        json.dump(data, f)
+
+
+def load_json(filename):
+    """Load data from a JSON file."""
+    with open(filename, encoding="utf8") as f:
+        return json.load(f)
+
+
+def save_csv(filename, data, header=""):
+    """Save data as a CSV file."""
+    np.savetxt(
+        filename, data, fmt="%d", delimiter=",", header=header, comments=""
+    )
+
+
+def load_csv(filename, skiprows=1):
+    """Load data from a CSV file."""
+    return np.loadtxt(filename, dtype=int, delimiter=",", skiprows=skiprows)
+
+
+def ignore_exceptions(func):
+    """Decorator that ignores all errors and warnings."""
+
+    def inner(*args, **kwargs):
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            try:
+                return func(*args, **kwargs)
+            except Exception:
+                return None
+
+    return inner
+
+
+def resolve_paths(func):
+    """Decorator that resolves all paths."""
+
+    def inner(*args, **kwargs):
+        parsed = func(*args, **kwargs)
+        for key in vars(parsed).keys():
+            if isinstance(getattr(parsed, key), pathlib.Path):
+                setattr(
+                    parsed, key, getattr(parsed, key).expanduser().resolve()
+                )
+        return parsed
+
+    return inner

model.json

@@ -0,0 +1,11 @@
+{
+  "name": "MMT",
+  "package_dir": "mmt",
+  "entry_point": ["representation", "music_x_transformers"],
+  "checkpoint": {
+    "location": "committed directly to Space repo at checkpoints/sod-ape/",
+    "structure": "train-args.json + checkpoints/best_model.pt, preserving the exact relative layout mmt/generate.py expects (train_args = utils.load_json(out_dir / 'train-args.json'); model.load_state_dict(torch.load(out_dir / 'checkpoints' / 'best_model.pt'))",
+    "size_mb": 80,
+    "source": "sod/ape/ from the README's Pretrained Models SharePoint link"
+  }
+}

packages.txt

@@ -0,0 +1 @@
+fluidsynth

requirements.txt

@@ -0,0 +1,9 @@
+git+https://github.com/TEAMuP-dev/pyharp.git@develop
+# model-specific deps below:
+torch==2.8.0
+torchaudio==2.8.0
+x-transformers==0.25.9
+einops
+muspy
+pretty_midi
+numpy