SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2310.16334 · NeurIPS · 2024

Structured Multi-Track Accompaniment Arrangement via Style Prior Modelling

Ziyu Wang, Ye Wang, Gus Xia, Jingwei Zhao

arXiv · PDF · Open in the Atlas

Code that ran

13 of 16 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 13 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2310.16334.json · badge: /api/ran/2310.16334.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
zhaojw1998/accomontage-3 canonical (classifier label)
classifier basis: census
13 ran · 3 no recorded run
FunctionStatusWhere it lives
accumulate_loss_dic Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/train_prior.py
code served (permissive licence) · get_code("5bd03b0ee49ce8ff")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch import Tensor

torch.manual_seed(0)

# Create a mock model that returns a scalar loss tensor
class MockModel(nn.Module):
    def forward(self, mode, *args):
        if mode == 'loss':
            return torch.tensor(0.5, requires_grad=True)
        return None

# Create mock batch data
mock_batch_x = torch.randn(2, 4)
mock_batch_y = torch.randn(2, 3)

# Create a simple dataloader (list of batches)
dataloader = [(mock_batch_x, mock_batch_y)]

# Create a mock summary writer
class MockSummaryWriter:
    def write_task(self, task_name, loss_dict, epoch):
        pass

summary_writers = MockSummaryWriter()

# Set up the ARGS for accumulate_loss_dic function
ARGS = {
    'writer_names': ['func_l'],
    'loss_dic': {'func_l': 0.0},
    'loss_items': [torch.tensor(0.25, requires_grad=True)]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collate_fn Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/orchestrator/autoencoder_dataset.py
code served (permissive licence) · get_code("b6a18c9188687399")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch

torch.manual_seed(0)

# Create small batch items
# Each item is: (pr, programs, _, _)
# pr shape: (num_tracks, 32, 128) - pianoroll matrices
# programs shape: (num_tracks,) - instrument indices

batch_item_1 = (
    np.random.randint(0, 10, size=(2, 32, 128), dtype=np.int32),  # pr: 2 tracks, 32 time steps, 128 pitches
    np.array([0, 5], dtype=np.int32),  # programs: 2 instruments
    None,
    None
)

batch_item_2 = (
    np.random.randint(0, 10, size=(1, 32, 128), dtype=np.int32),  # pr: 1 track
    np.array([10], dtype=np.int32),  # programs: 1 instrument
    None,
    None
)

batch = [batch_item_1, batch_item_2]

device = torch.device('cpu')

ARGS = {
    'batch': batch,
    'device': device,
    'pitch_shift': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collate_fn_inference Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/orchestrator/autoencoder_dataset.py
code served (permissive licence) · get_code("eb2c51653d85632a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch

torch.manual_seed(0)

# Create minimal batch data for collate_fn_inference
# batch is expected to be a list with 1 element
# Each element is (tracks, instrument, dynamics, song_dir)

# tracks: shape (track, time, 128) - piano roll format
# The function expects time % 32 == 0 or will pad it
# Let's create: 2 tracks, 64 time steps, 128 pitches
tracks = np.random.randint(0, 5, size=(2, 64, 128), dtype=np.int32).astype(np.float32)

# instrument: shape (track, num_instruments) - will be repeated
# Let's use shape (2, 4) for 2 tracks, 4 instrument classes
instrument = np.random.randint(0, 10, size=(2, 4), dtype=np.int32).astype(np.float32)

# dynamics: optional, can be None or shape (track, time, pitches, features)
# Let's set to None to keep it simple
dynamics = None

# song_dir: just a string identifier
song_dir = "test_song"

# Create the batch as a list with one tuple
batch = [(tracks, instrument, dynamics, song_dir)]

# device: CPU device
device = torch.device('cpu')

ARGS = {
    'batch': batch,
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_center_pitch Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/data_preprocessing/lmd_vq_compression.py
code served (permissive licence) · get_code("1ab84f366812f8de")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Analyze the function:
# - pr is indexed with axis=(-2, -1), so it needs at least 3 dimensions
# - np.arange(0, 128) is used, suggesting pr's last dimension should be 128 (pitch dimension)
# - The function sums over the last two dimensions, leaving earlier dimensions intact
# - pr appears to be a piano roll representation: shape should be (batch, time, pitch) or similar

# Create a small valid pr array
# Shape: (batch=2, time=4, pitch=128) - minimal but valid
pr = np.random.rand(2, 4, 128).astype(np.float32)

# Ensure some non-zero values so the function works properly
pr = pr * 0.5 + 0.1  # scale to have reasonable values

ARGS = {
    "pr": pr
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_pr_feat Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/data_preprocessing/lmd_vq_compression.py
code served (permissive licence) · get_code("787065ce7bbdf5f9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function compute_pr_feat takes a single parameter 'pr'
# Looking at the function body:
# - pr is used with np.sum(pr, axis=-1) which suggests it's a numpy array
# - pr is compared with > 0, suggesting numeric values
# - The axis=-1 operation suggests it's at least 2D
# - The division by 14 and clipping suggests values are numeric
# - pr should have shape that makes sense for pitch roll (pr)

# Creating a small numpy array with shape (2, 14)
# - batch size 2 (distinct from other dims)
# - 14 pitch classes (used in the division by 14)
pr = np.random.rand(2, 14) * 2  # Random values between 0 and 2 to have some zeros and some non-zeros

ARGS = {
    "pr": pr
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
convert_pop909 Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhaojw1998/accomontage-3/data_preprocessing/pop909_process_4bin_data.py
code served (permissive licence) · get_code("aafb89990a0777de")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 4, dtype family float, shape (3, 12, 128, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Analyze the function:
# - melody, bridge, piano are iterables of tuples with 8 elements each: (sb, sq, sde, eb, eq, ede, p, v)
# - beat is used only for its length: len(beat) * 4 determines output shape
# - Each tuple unpacks to 8 values; sde and ede are asserted to be 4
# - sb, sq, eb, eq are used in arithmetic; p is converted to int; v is a velocity value

# Create small valid data structures
beat = [0, 1, 2]  # len(beat) = 3, so output will have shape (3, 12, 128, 2)

# Each entry: (sb, sq, sde, eb, eq, ede, p, v)
# sde and ede must be 4 (asserted)
# sb, sq: bar and quarter position
# eb, eq: end bar and quarter position
# p: pitch (0-127)
# v: velocity

melody = [
    (0, 0, 4, 0, 2, 4, 60, 0.8),
    (0, 2, 4, 1, 0, 4, 64, 0.9),
]

bridge = [
    (1, 0, 4, 1, 3, 4, 62, 0.7),
]

piano = [
    (0, 1, 4, 0, 3, 4, 48, 0.6),
    (1, 1, 4, 2, 0, 4, 50, 0.75),
]

ARGS = {
    "melody": melody,
    "bridge": bridge,
    "piano": piano,
    "beat": beat,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (3, 12, 128, 2) — shape (3, 12, 128, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
init_loss_dic Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/train_autoencoder.py
code served (permissive licence) · get_code("c08587ef6d2aa297")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    'writer_names': ['loss', 'pno_tree_l', 'pl', 'dl', 'kl_l', 'kl_pno_red', 'kl_trf', 'feat_l', 'onset_l', 'intensity_l', 'center_l', 'vqvae_l', 'func_l', 'cmt_l', 'plty']
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
init_loss_dic Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/train_prior.py
code served (permissive licence) · get_code("1e6b663103a98f7d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function iterates over writer_names with a for loop and uses each key
# to initialize loss_dic[key] = 0.0
# So writer_names should be an iterable of strings (keys)

writer_names = ["train_loss", "val_loss", "test_loss"]

ARGS = {
    "writer_names": writer_names
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
midi2matrix Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/data_preprocessing/lmd_midi_quantization.py
code served (permissive licence) · get_code("8cec2ae9c2607f34")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

class MockNote:
    def __init__(self, start, end, pitch):
        self.start = start
        self.end = end
        self.pitch = pitch

class MockInstrument:
    def __init__(self, program, is_drum=False, notes=None):
        self.program = program
        self.is_drum = is_drum
        self.notes = notes if notes is not None else []

class MockMidi:
    def __init__(self, instruments=None):
        self.instruments = instruments if instruments is not None else []

# Create a quaver array (time quantization grid)
quaver = np.array([0.0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0])

# Create mock notes
notes_track1 = [
    MockNote(start=0.1, end=1.2, pitch=60),
    MockNote(start=1.5, end=2.8, pitch=64),
]

notes_track2 = [
    MockNote(start=0.3, end=1.0, pitch=67),
]

# Create mock instruments (skip drum tracks)
instrument1 = MockInstrument(program=0, is_drum=False, notes=notes_track1)
instrument2 = MockInstrument(program=33, is_drum=False, notes=notes_track2)
drum_track = MockInstrument(program=128, is_drum=True, notes=[])

# Create mock MIDI object
midi = MockMidi(instruments=[instrument1, drum_track, instrument2])

ARGS = {
    "midi": midi,
    "quaver": quaver,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
midi2matrix Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/data_preprocessing/pop909_process_4bin_data.py
code served (permissive licence) · get_code("b1539088e28cabcb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a minimal MIDI-like object structure
class Note:
    def __init__(self, start, end, pitch, velocity):
        self.start = start
        self.end = end
        self.pitch = pitch
        self.velocity = velocity

class ControlChange:
    def __init__(self, time, number, value):
        self.time = time
        self.number = number
        self.value = value

class Instrument:
    def __init__(self, program, notes, control_changes):
        self.program = program
        self.notes = notes
        self.control_changes = control_changes

class MIDIFile:
    def __init__(self, instruments):
        self.instruments = instruments

# Create quaver time grid
quaver = np.array([0.0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0])

# Create notes for first instrument
notes_1 = [
    Note(start=0.0, end=1.0, pitch=60, velocity=100),
    Note(start=1.0, end=2.0, pitch=64, velocity=90),
    Note(start=2.5, end=3.5, pitch=67, velocity=85),
]

# Create control changes for first instrument
control_changes_1 = [
    ControlChange(time=0.0, number=7, value=100),
    ControlChange(time=2.0, number=11, value=64),
]

# Create notes for second instrument
notes_2 = [
    Note(start=0.5, end=1.5, pitch=48, velocity=95),
    Note(start=2.0, end=3.0, pitch=52, velocity=88),
]

# Create control changes for second instrument
control_changes_2 = [
    ControlChange(time=1.0, number=7, value=90),
]

# Create instruments
instrument_1 = Instrument(program=0, notes=notes_1, control_changes=control_changes_1)
instrument_2 = Instrument(program=33, notes=notes_2, control_changes=control_changes_2)

# Create MIDI file
midi = MIDIFile(instruments=[instrument_1, instrument_2])

# Define arguments
ARGS = {
    "midi": midi,
    "quaver": quaver,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
midi2matrix Ran
ran; our contract draft was wrong, not the code
zhaojw1998/accomontage-3/data_preprocessing/slakh_midi_quantization.py
code served (permissive licence) · get_code("f3e3d404d14955dc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

class Note:
    def __init__(self, start, end, pitch):
        self.start = start
        self.end = end
        self.pitch = pitch

class Instrument:
    def __init__(self, program, notes):
        self.program = program
        self.notes = notes

class MidiFile:
    def __init__(self, instruments):
        self.instruments = instruments

# Create a quaver (time quantization grid)
quaver = np.array([0.0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0])

# Create some notes for a track
notes_track1 = [
    Note(start=0.0, end=1.0, pitch=60),
    Note(start=1.0, end=2.0, pitch=64),
    Note(start=2.0, end=3.5, pitch=67),
]

notes_track2 = [
    Note(start=0.5, end=1.5, pitch=48),
    Note(start=2.5, end=3.5, pitch=52),
]

# Create instruments
instrument1 = Instrument(program=0, notes=notes_track1)
instrument2 = Instrument(program=33, notes=notes_track2)

# Create MIDI file with instruments
midi = MidiFile(instruments=[instrument1, instrument2])

ARGS = {
    "midi": midi,
    "quaver": quaver,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
pr_mat_pitch_shift Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhaojw1998/accomontage-3/orchestrator/autoencoder_dataset.py
code served (permissive licence) · get_code("f566bcd9030d470c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

np.random.seed(0)

# pr_mat_pitch_shift takes:
# - pr_mat: a numpy array (pitch roll matrix)
# - shift: an integer (number of positions to roll)

# Create a small pitch roll matrix - shape (2, 4) where rows are time/samples and columns are pitch classes
pr_mat = np.random.rand(2, 4).astype(np.float32)

# shift: integer for rolling along the last axis
shift = 2

ARGS = {
    "pr_mat": pr_mat,
    "shift": shift,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
write_loss_to_dic Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhaojw1998/accomontage-3/train_prior.py
code served (permissive licence) · get_code("a89b5c46d7a86b60")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create small test tensors with .item() method
loss_tensor_1 = torch.tensor(0.5, requires_grad=False)
loss_tensor_2 = torch.tensor(0.3, requires_grad=False)
loss_tensor_3 = torch.tensor(0.7, requires_grad=False)

# Define writer_names to match the number of loss items
writer_names = ['func_l', 'reg_l', 'total_l']

# Create loss_items as list of tensors
loss_items = [loss_tensor_1, loss_tensor_2, loss_tensor_3]

ARGS = {
    'writer_names': writer_names,
    'loss_items': loss_items,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_lead_sheet Not yet run
unverified (harvested, no recorded run)
zhaojw1998/accomontage-3/piano_arranger/AccoMontage.py
code served (permissive licence) · get_code("062edc011308d901")
How it ran

no invocation recorded.

phrase_selection Not yet run
unverified (harvested, no recorded run)
zhaojw1998/accomontage-3/piano_arranger/AccoMontage.py
code served (permissive licence) · get_code("b8c51d7614d3def8")
How it ran

no invocation recorded.

set_premises Not yet run
unverified (harvested, no recorded run)
zhaojw1998/accomontage-3/piano_arranger/AccoMontage.py
code served (permissive licence) · get_code("12474652ed2e74df")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 0 later) and 13 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
phrase_selection 1 other paper
1 dated earlier
2108.11213 (2021-08)
AccoMontage: Accompaniment Arrangement via Phrase Selection and Style Transfer
set_premises 1 other paper
1 dated earlier
2108.11213 (2021-08)
AccoMontage: Accompaniment Arrangement via Phrase Selection and Style Transfer

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

In the realm of music AI, arranging rich and structured multi-track accompaniments from a simple lead sheet presents significant challenges. Such challenges include maintaining track cohesion, ensuring long-term coherence, and optimizing computational efficiency. In this paper, we introduce a novel system that leverages prior modelling over disentangled style factors to address these challenges. Our method presents a two-stage process: initially, a piano arrangement is derived from the lead sheet by retrieving piano texture styles; subsequently, a multi-track orchestration is generated by infusing orchestral function styles into the piano arrangement. Our key design is the use of vector quantization and a unique multistream Transformer to model the long-term flow of the orchestration style, which enables flexible, controllable, and structured music generation. Experiments show that by factorizing the arrangement task into interpretable sub-stages, our approach enhances generative capacity while improving efficiency. Additionally, our system supports a variety of music genres and provides style control at different composition hierarchies. We further show that our system achieves superior coherence, structure, and overall arrangement quality compared to existing baselines.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2310.16334")
get_code_for_paper("2310.16334")
have("2310.16334")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2310.16334.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2310.16334.svg)](https://syntology.ai/paper/2310.16334)

Connect an agent — have() is free.