Skip to content

GitLab

Unverified Commit 61a51f5f authored by Arthur's avatar Arthur Committed by GitHub
Browse files

Add Jukebox model (replaces #16875) (#17826)

parent 9740a03f
Hide whitespace changes
Inline Side-by-side
Showing with 557 additions and 0 deletions
tests/models/jukebox/test_modeling_jukebox.py 0 → 100644
View file @ 61a51f5f
# coding=utf-8
# Copyright 2022 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow
from transformers.trainer_utils import set_seed
if is_torch_available():
import torch
from transformers import JukeboxModel, JukeboxTokenizer
@require_torch
class Jukebox1bModelTester(unittest.TestCase):
all_model_classes = (JukeboxModel,) if is_torch_available() else ()
model_id = "openai/jukebox-1b-lyrics"
metas = dict(
artist="Zac Brown Band",
genres="Country",
lyrics="""I met a traveller from an antique land,
Who said "Two vast and trunkless legs of stone
Stand in the desert. . . . Near them, on the sand,
Half sunk a shattered visage lies, whose frown,
And wrinkled lip, and sneer of cold command,
Tell that its sculptor well those passions read
Which yet survive, stamped on these lifeless things,
The hand that mocked them, and the heart that fed;
And on the pedestal, these words appear:
My name is Ozymandias, King of Kings;
Look on my Works, ye Mighty, and despair!
Nothing beside remains. Round the decay
Of that colossal Wreck, boundless and bare
The lone and level sands stretch far away
""",
)
# fmt: off
EXPECTED_OUTPUT_2 = [
1864, 1536, 1213, 1870, 1357, 1536, 519, 880, 1323, 789, 1082, 534,
1000, 1445, 1105, 1130, 967, 515, 1434, 1620, 534, 1495, 283, 1445,
333, 1307, 539, 1631, 1528, 375, 1434, 673, 627, 710, 778, 1883,
1405, 1276, 1455, 1228
]
EXPECTED_OUTPUT_1 = [
1125, 1751, 697, 1776, 1141, 1476, 391, 697, 1125, 684, 867, 416,
844, 1372, 1274, 717, 1274, 844, 1299, 1419, 697, 1370, 317, 1125,
191, 1440, 1370, 1440, 1370, 282, 1621, 1370, 368, 349, 867, 1872,
1262, 869, 1728, 747
]
EXPECTED_OUTPUT_0 = [
1755, 842, 307, 1843, 1022, 1395, 234, 1554, 806, 739, 1022, 442,
616, 556, 268, 1499, 933, 457, 1440, 1837, 755, 985, 308, 902,
293, 1443, 1671, 1141, 1533, 555, 1562, 1061, 287, 417, 1022, 2008,
1186, 1015, 1777, 268
]
EXPECTED_Y_COND = [1058304, 0, 786432, 7169, 507, 76, 27, 40, 30, 76]
EXPECTED_PRIMED_0 = [
390, 1160, 1002, 1907, 1788, 1788, 1788, 1907, 1002, 1002, 1854, 1002,
1002, 1002, 1002, 1002, 1002, 1160, 1160, 1606, 596, 596, 1160, 1002,
1516, 596, 1002, 1002, 1002, 1907, 1788, 1788, 1788, 1854, 1788, 1907,
1907, 1788, 596, 1626
]
EXPECTED_PRIMED_1 = [
1236, 1668, 1484, 1920, 1848, 1409, 139, 864, 1828, 1272, 1599, 824,
1672, 139, 555, 1484, 824, 1920, 555, 596, 1579, 1599, 1231, 1599,
1637, 1407, 212, 824, 1599, 116, 1433, 824, 258, 1599, 1433, 1895,
1063, 1433, 1433, 1599
]
EXPECTED_PRIMED_2 = [
1684, 1873, 1119, 1189, 395, 611, 1901, 972, 890, 1337, 1392, 1927,
96, 972, 672, 780, 1119, 890, 158, 771, 1073, 1927, 353, 1331,
1269, 1459, 1333, 1645, 812, 1577, 1337, 606, 353, 981, 1466, 619,
197, 391, 302, 1930
]
EXPECTED_VQVAE_ENCODE = [
390, 1160, 1002, 1907, 1788, 1788, 1788, 1907, 1002, 1002, 1854, 1002,
1002, 1002, 1002, 1002, 1002, 1160, 1160, 1606, 596, 596, 1160, 1002,
1516, 596, 1002, 1002, 1002, 1907, 1788, 1788, 1788, 1854, 1788, 1907,
1907, 1788, 596, 1626
]
EXPECTED_VQVAE_DECODE = [
-0.0492, -0.0524, -0.0565, -0.0640, -0.0686, -0.0684, -0.0677, -0.0664,
-0.0605, -0.0490, -0.0330, -0.0168, -0.0083, -0.0075, -0.0051, 0.0025,
0.0136, 0.0261, 0.0386, 0.0497, 0.0580, 0.0599, 0.0583, 0.0614,
0.0740, 0.0889, 0.1023, 0.1162, 0.1211, 0.1212, 0.1251, 0.1336,
0.1502, 0.1686, 0.1883, 0.2148, 0.2363, 0.2458, 0.2507, 0.2531
]
EXPECTED_AUDIO_COND = [
0.0256, -0.0544, 0.1600, -0.0032, 0.1066, 0.0825, -0.0013, 0.3440,
0.0210, 0.0412, -0.1777, -0.0892, -0.0164, 0.0285, -0.0613, -0.0617,
-0.0137, -0.0201, -0.0175, 0.0215, -0.0627, 0.0520, -0.0730, 0.0970,
-0.0100, 0.0442, -0.0586, 0.0207, -0.0015, -0.0082
]
EXPECTED_META_COND = [
0.0415, 0.0877, 0.0022, -0.0055, 0.0751, 0.0334, 0.0324, -0.0068,
0.0011, 0.0017, -0.0676, 0.0655, -0.0143, 0.0399, 0.0303, 0.0743,
-0.0168, -0.0394, -0.1113, 0.0124, 0.0442, 0.0267, -0.0003, -0.1536,
-0.0116, -0.1837, -0.0180, -0.1026, -0.0777, -0.0456
]
EXPECTED_LYRIC_COND = [
76, 27, 40, 30, 76, 46, 44, 47, 40, 37, 38, 31, 45, 45, 76, 38, 31, 33,
45, 76, 41, 32, 76, 45, 46, 41, 40, 31, 78, 76
]
# fmt: on
def prepare_inputs(self):
tokenizer = JukeboxTokenizer.from_pretrained(self.model_id)
tokens = tokenizer(**self.metas)["input_ids"]
return tokens
@slow
def test_sampling(self):
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval()
labels = self.prepare_inputs()
set_seed(0)
zs = [torch.zeros(1, 0, dtype=torch.long).cpu() for _ in range(3)]
zs = model._sample(zs, labels, [0], sample_length=40 * model.priors[0].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[0][0], torch.tensor(self.EXPECTED_OUTPUT_2))
set_seed(0)
zs = model._sample(zs, labels, [1], sample_length=40 * model.priors[1].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[1][0], torch.tensor(self.EXPECTED_OUTPUT_1))
set_seed(0)
zs = model._sample(zs, labels, [2], sample_length=40 * model.priors[2].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[2][0], torch.tensor(self.EXPECTED_OUTPUT_0))
@slow
def test_conditioning(self):
torch.backends.cuda.matmul.allow_tf32 = False
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval()
labels = self.prepare_inputs()
set_seed(0)
zs = [torch.zeros(1, 0, dtype=torch.long) for _ in range(3)]
top_prior = model.priors[0]
start = 0
music_token_conds = top_prior.get_music_tokens_conds(zs, start=start, end=start + top_prior.n_ctx)
metadata = top_prior.get_metadata(labels[0].clone(), start, 1058304, 0)
self.assertIsNone(music_token_conds)
self.assertListEqual(metadata.numpy()[0][:10].tolist(), self.EXPECTED_Y_COND)
audio_conditioning, metadata_conditioning, lyric_tokens = top_prior.get_cond(music_token_conds, metadata)
torch.testing.assert_allclose(
audio_conditioning[0][0][:30].detach(), torch.tensor(self.EXPECTED_AUDIO_COND), atol=1e-4, rtol=1e-4
)
torch.testing.assert_allclose(
metadata_conditioning[0][0][:30].detach(), torch.tensor(self.EXPECTED_META_COND), atol=1e-4, rtol=1e-4
)
torch.testing.assert_allclose(
lyric_tokens[0, :30].detach(), torch.tensor(self.EXPECTED_LYRIC_COND), atol=1e-4, rtol=1e-4
)
@slow
def test_primed_sampling(self):
torch.backends.cuda.matmul.allow_tf32 = False
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval()
set_seed(0)
waveform = torch.rand((1, 5120, 1))
tokens = [i for i in self.prepare_inputs()]
zs = [model.vqvae.encode(waveform, start_level=2, bs_chunks=waveform.shape[0])[0], None, None]
zs = model._sample(
zs, tokens, sample_levels=[0], save_results=False, sample_length=40 * model.priors[0].raw_to_tokens
)
torch.testing.assert_allclose(zs[0][0][:40], torch.tensor(self.EXPECTED_PRIMED_0))
upper_2 = torch.cat((zs[0], torch.zeros(1, 2048 - zs[0].shape[-1])), dim=-1).long()
zs = [upper_2, model.vqvae.encode(waveform, start_level=1, bs_chunks=waveform.shape[0])[0], None]
zs = model._sample(
zs, tokens, sample_levels=[1], save_results=False, sample_length=40 * model.priors[1].raw_to_tokens
)
torch.testing.assert_allclose(zs[1][0][:40], torch.tensor(self.EXPECTED_PRIMED_1))
upper_1 = torch.cat((zs[1], torch.zeros(1, 2048 - zs[1].shape[-1])), dim=-1).long()
zs = [upper_2, upper_1, model.vqvae.encode(waveform, start_level=0, bs_chunks=waveform.shape[0])[0]]
zs = model._sample(
zs, tokens, sample_levels=[2], save_results=False, sample_length=40 * model.priors[2].raw_to_tokens
)
torch.testing.assert_allclose(zs[2][0][:40].cpu(), torch.tensor(self.EXPECTED_PRIMED_2))
@slow
def test_vqvae(self):
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval()
set_seed(0)
x = torch.rand((1, 5120, 1))
with torch.no_grad():
zs = model.vqvae.encode(x, start_level=2, bs_chunks=x.shape[0])
torch.testing.assert_allclose(zs[0][0], torch.tensor(self.EXPECTED_VQVAE_ENCODE))
with torch.no_grad():
x = model.vqvae.decode(zs, start_level=2, bs_chunks=x.shape[0])
torch.testing.assert_allclose(x[0, :40, 0], torch.tensor(self.EXPECTED_VQVAE_DECODE), atol=1e-4, rtol=1e-4)
@require_torch
class Jukebox5bModelTester(unittest.TestCase):
all_model_classes = (JukeboxModel,) if is_torch_available() else ()
model_id = "openai/jukebox-5b-lyrics"
metas = dict(
artist="Zac Brown Band",
genres="Country",
lyrics="""I met a traveller from an antique land,
Who said "Two vast and trunkless legs of stone
Stand in the desert. . . . Near them, on the sand,
Half sunk a shattered visage lies, whose frown,
And wrinkled lip, and sneer of cold command,
Tell that its sculptor well those passions read
Which yet survive, stamped on these lifeless things,
The hand that mocked them, and the heart that fed;
And on the pedestal, these words appear:
My name is Ozymandias, King of Kings;
Look on my Works, ye Mighty, and despair!
Nothing beside remains. Round the decay
Of that colossal Wreck, boundless and bare
The lone and level sands stretch far away
""",
)
# fmt: off
EXPECTED_OUTPUT_2 = [
1489, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
1489, 1489, 1489, 1489, 1150, 1853, 1509, 1150, 1357, 1509, 6, 1272
]
EXPECTED_OUTPUT_1 = [
1125, 416, 1125, 1125, 1125, 1125, 1125, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416
]
EXPECTED_OUTPUT_0 = [
1755, 1061, 234, 1755, 1061, 1755, 185, 290, 307, 307, 616, 616,
616, 616, 616, 616, 307, 290, 417, 1755, 234, 1755, 185, 290,
290, 290, 307, 616, 616, 616, 616, 616, 290, 234, 234, 1755,
234, 234, 1755, 234, 185, 185, 307, 616, 616, 616, 616, 290,
1755, 1755, 1755, 234, 234, 1755, 1572, 290, 307, 616, 34, 616
]
EXPECTED_GPU_OUTPUTS_2 = [
1489, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653,
653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653, 653
]
EXPECTED_GPU_OUTPUTS_1 = [
1125, 1125, 416, 1125, 1125, 416, 1125, 1125, 416, 416, 1125, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416,
416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416, 416
]
EXPECTED_GPU_OUTPUTS_0 = [
491, 1755, 34, 1613, 1755, 417, 992, 1613, 222, 842, 1353, 1613,
844, 632, 185, 1613, 844, 632, 185, 1613, 185, 842, 677, 1613,
185, 114, 1353, 1613, 307, 89, 844, 1613, 307, 1332, 234, 1979,
307, 89, 1353, 616, 34, 842, 185, 842, 34, 842, 185, 842,
307, 114, 185, 89, 34, 1268, 185, 89, 34, 842, 185, 89
]
# fmt: on
def prepare_inputs(self, model_id):
tokenizer = JukeboxTokenizer.from_pretrained(model_id)
tokens = tokenizer(**self.metas)["input_ids"]
return tokens
@slow
def test_sampling(self):
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval()
labels = self.prepare_inputs(self.model_id)
set_seed(0)
zs = [torch.zeros(1, 0, dtype=torch.long).cpu() for _ in range(3)]
zs = model._sample(zs, labels, [0], sample_length=60 * model.priors[0].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[0][0], torch.tensor(self.EXPECTED_OUTPUT_2))
set_seed(0)
zs = model._sample(zs, labels, [1], sample_length=60 * model.priors[1].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[1][0], torch.tensor(self.EXPECTED_OUTPUT_1))
set_seed(0)
zs = model._sample(zs, labels, [2], sample_length=60 * model.priors[2].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[2][0], torch.tensor(self.EXPECTED_OUTPUT_0))
@slow
def test_slow_sampling(self):
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval().to("cuda")
labels = [i.cuda() for i in self.prepare_inputs(self.model_id)]
set_seed(0)
model.priors[0].cuda()
zs = [torch.zeros(1, 0, dtype=torch.long).cuda() for _ in range(3)]
zs = model._sample(zs, labels, [0], sample_length=60 * model.priors[0].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[0][0].cpu(), torch.tensor(self.EXPECTED_GPU_OUTPUTS_2))
model.priors[0].cpu()
set_seed(0)
model.priors[1].cuda()
zs = model._sample(zs, labels, [1], sample_length=60 * model.priors[1].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[1][0].cpu(), torch.tensor(self.EXPECTED_GPU_OUTPUTS_1))
model.priors[1].cpu()
set_seed(0)
model.priors[2].cuda()
zs = model._sample(zs, labels, [2], sample_length=60 * model.priors[2].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[2][0].cpu(), torch.tensor(self.EXPECTED_GPU_OUTPUTS_0))
@slow
def test_fp16_slow_sampling(self):
model = JukeboxModel.from_pretrained(self.model_id, min_duration=0).eval().half().to("cuda")
labels = [i.cuda() for i in self.prepare_inputs(self.model_id)]
set_seed(0)
zs = [torch.zeros(1, 0, dtype=torch.long).cuda() for _ in range(3)]
zs = model._sample(zs, labels, [0], sample_length=60 * model.priors[0].raw_to_tokens, save_results=False)
torch.testing.assert_allclose(zs[0][0].cpu(), torch.tensor(self.EXPECTED_GPU_OUTPUTS_2))
tests/models/jukebox/test_tokenization_jukebox.py 0 → 100644
View file @ 61a51f5f
# coding=utf-8
# Copyright 2022 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
from transformers import JukeboxTokenizer
from transformers.testing_utils import require_torch
class JukeboxTokenizationTest(unittest.TestCase):
tokenizer_class = JukeboxTokenizer
metas = dict(
artist="Zac Brown Band",
genres="Country",
lyrics="""I met a traveller from an antique land,
Who said "Two vast and trunkless legs of stone
Stand in the desert. . . . Near them, on the sand,
Half sunk a shattered visage lies, whose frown,
And wrinkled lip, and sneer of cold command,
Tell that its sculptor well those passions read
Which yet survive, stamped on these lifeless things,
The hand that mocked them, and the heart that fed;
And on the pedestal, these words appear:
My name is Ozymandias, King of Kings;
Look on my Works, ye Mighty, and despair!
Nothing beside remains. Round the decay
Of that colossal Wreck, boundless and bare
The lone and level sands stretch far away
""",
)
@require_torch
def test_1b_lyrics_tokenizer(self):
"""
how to run the same test with openAI
...
"""
import torch
tokenizer = JukeboxTokenizer.from_pretrained("openai/jukebox-1b-lyrics")
tokens = tokenizer(**self.metas)["input_ids"]
# fmt: off
EXPECTED_OUTPUT = [
torch.tensor([[
0, 0, 0, 7169, 507, 9, 76, 39, 31, 46, 76, 27,
76, 46, 44, 27, 48, 31, 38, 38, 31, 44, 76, 32,
44, 41, 39, 76, 27, 40, 76, 27, 40, 46, 35, 43,
47, 31, 76, 38, 27, 40, 30, 64, 78, 76, 76, 76,
76, 76, 76, 76, 76, 23, 34, 41, 76, 45, 27, 35,
30, 76, 71, 20, 49, 41, 76, 48, 27, 45, 46, 76,
27, 40, 30, 76, 46, 44, 47, 40, 37, 38, 31, 45,
45, 76, 38, 31, 33, 45, 76, 41, 32, 76, 45, 46,
41, 40, 31, 78, 76, 76, 76, 76, 76, 76, 76, 76,
19, 46, 27, 40, 30, 76, 35, 40, 76, 46, 34, 31,
76, 30, 31, 45, 31, 44, 46, 63, 76, 63, 76, 63,
76, 63, 76, 14, 31, 27, 44, 76, 46, 34, 31, 39,
64, 76, 41, 40, 76, 46, 34, 31, 76, 45, 27, 40,
30, 64, 78, 76, 76, 76, 76, 76, 76, 76, 76, 8,
27, 38, 32, 76, 45, 47, 40, 37, 76, 27, 76, 45,
34, 27, 46, 46, 31, 44, 31, 30, 76, 48, 35, 45,
27, 33, 31, 76, 38, 35, 31, 45, 64, 76, 49, 34,
41, 45, 31, 76, 32, 44, 41, 49, 40, 64, 78, 76,
76, 76, 76, 76, 76, 76, 76, 1, 40, 30, 76, 49,
44, 35, 40, 37, 38, 31, 30, 76, 38, 35, 42, 64,
76, 27, 40, 30, 76, 45, 40, 31, 31, 44, 76, 41,
32, 76, 29, 41, 38, 30, 76, 29, 41, 39, 39, 27,
40, 30, 64, 78, 76, 76, 76, 76, 76, 76, 76, 76,
20, 31, 38, 38, 76, 46, 34, 27, 46, 76, 35, 46,
45, 76, 45, 29, 47, 38, 42, 46, 41, 44, 76, 49,
31, 38, 38, 76, 46, 34, 41, 45, 31, 76, 42, 27,
45, 45, 35, 41, 40, 45, 76, 44, 31, 27, 30, 78,
76, 76, 76, 76, 76, 76, 76, 76, 23, 34, 35, 29,
34, 76, 51, 31, 46, 76, 45, 47, 44, 48, 35, 48,
31, 64, 76, 45, 46, 27, 39, 42, 31, 30, 76, 41,
40, 76, 46, 34, 31, 45, 31, 76, 38, 35, 32, 31,
38, 31, 45, 45, 76, 46, 34, 35, 40, 33, 45, 64,
78, 76, 76, 76, 76, 76, 76, 76, 76, 20, 34, 31,
76, 34, 27, 40, 30, 76, 46, 34, 27, 46, 76, 39,
41, 29, 37, 31, 30, 76, 46, 34, 31, 39, 64, 76,
27, 40, 30, 76, 46, 34, 31, 76, 34, 31, 27, 44,
46, 76, 46, 34, 27, 46, 76, 32, 31, 30, 66, 78,
76, 76, 76, 76, 76, 76, 76, 76, 1, 40, 30, 76,
41, 40, 76, 46, 34, 31, 76, 42, 31, 30, 31, 45,
46, 27, 38, 64, 76, 46, 34, 31, 45, 31, 76, 49,
41, 44, 30, 45, 76, 27, 42, 42, 31, 27, 44, 65,
78, 76, 76, 76, 76, 76, 76, 76, 76, 13, 51, 76,
40, 27, 39, 31, 76, 35, 45, 76, 15, 52, 51, 39,
27, 40, 30, 35, 27, 45, 64, 76, 11, 35, 40, 33,
76, 41, 32, 76, 11, 35, 40, 33, 45, 66, 78, 76,
76, 76, 76, 76, 76, 76, 76, 12, 41, 41, 37, 76,
41, 40, 76, 39, 51, 76, 23, 41, 44, 37, 45, 64,
76, 51, 31, 76, 13, 35, 33, 34, 46, 51, 64, 76,
27, 40, 30, 76, 30, 31, 45, 42, 27, 35, 44, 67,
78, 76, 76, 76, 76, 76, 76, 76, 76, 14, 41, 46,
34, 35, 40, 33, 76, 28, 31, 45, 35, 30, 31, 76,
44, 31, 39, 27, 35, 40, 45, 63, 76, 18, 41, 47,
40, 30, 76, 46, 34, 31, 76, 30, 31, 29, 27, 51,
78, 76, 76, 76, 76, 76, 76, 76, 76, 15, 32, 76,
46, 34, 27, 46, 76, 29, 41, 38, 41, 45, 45, 27,
38, 76, 23, 44, 31, 29, 37, 64, 76, 28, 41, 47,
40, 30, 38, 31, 45, 45, 76, 27, 40, 30, 76, 28,
27, 44, 31, 78, 76, 76, 76, 76, 76, 76, 76, 76,
20, 34, 31, 76, 38, 41, 40, 31, 76, 27, 40, 30,
76, 38, 31, 48, 31, 38, 76, 45, 27, 40, 30, 45,
76, 45, 46, 44, 31, 46, 29, 34, 76, 32, 27, 44,
76, 27, 49, 27, 51, 78, 76, 76, 76, 76, 76, 76,
76, 76]]),
torch.tensor([[0, 0, 0, 1069, 11]]),
torch.tensor([[0, 0, 0, 1069, 11]]),
]
# fmt: on
self.assertTrue(torch.allclose(tokens[0], EXPECTED_OUTPUT[0]))
self.assertTrue(torch.allclose(tokens[1], EXPECTED_OUTPUT[1]))
self.assertTrue(torch.allclose(tokens[2], EXPECTED_OUTPUT[2]))
@require_torch
def test_5b_lyrics_tokenizer(self):
"""
The outputs are similar that open AI but do not have the same format as this one is adapted to the HF integration.
"""
import torch
tokenizer = JukeboxTokenizer.from_pretrained("openai/jukebox-5b-lyrics")
tokens = tokenizer(**self.metas)["input_ids"]
# fmt: off
EXPECTED_OUTPUT = [
torch.tensor([[
0, 0, 0, 1069, 11, -1, -1, -1, -1, 9, 77, 39,
31, 46, 77, 27, 77, 46, 44, 27, 48, 31, 38, 38,
31, 44, 77, 32, 44, 41, 39, 77, 27, 40, 77, 27,
40, 46, 35, 43, 47, 31, 77, 38, 27, 40, 30, 64,
79, 77, 77, 77, 77, 77, 77, 77, 77, 23, 34, 41,
77, 45, 27, 35, 30, 77, 72, 20, 49, 41, 77, 48,
27, 45, 46, 77, 27, 40, 30, 77, 46, 44, 47, 40,
37, 38, 31, 45, 45, 77, 38, 31, 33, 45, 77, 41,
32, 77, 45, 46, 41, 40, 31, 79, 77, 77, 77, 77,
77, 77, 77, 77, 19, 46, 27, 40, 30, 77, 35, 40,
77, 46, 34, 31, 77, 30, 31, 45, 31, 44, 46, 63,
77, 63, 77, 63, 77, 63, 77, 14, 31, 27, 44, 77,
46, 34, 31, 39, 64, 77, 41, 40, 77, 46, 34, 31,
77, 45, 27, 40, 30, 64, 79, 77, 77, 77, 77, 77,
77, 77, 77, 8, 27, 38, 32, 77, 45, 47, 40, 37,
77, 27, 77, 45, 34, 27, 46, 46, 31, 44, 31, 30,
77, 48, 35, 45, 27, 33, 31, 77, 38, 35, 31, 45,
64, 77, 49, 34, 41, 45, 31, 77, 32, 44, 41, 49,
40, 64, 79, 77, 77, 77, 77, 77, 77, 77, 77, 1,
40, 30, 77, 49, 44, 35, 40, 37, 38, 31, 30, 77,
38, 35, 42, 64, 77, 27, 40, 30, 77, 45, 40, 31,
31, 44, 77, 41, 32, 77, 29, 41, 38, 30, 77, 29,
41, 39, 39, 27, 40, 30, 64, 79, 77, 77, 77, 77,
77, 77, 77, 77, 20, 31, 38, 38, 77, 46, 34, 27,
46, 77, 35, 46, 45, 77, 45, 29, 47, 38, 42, 46,
41, 44, 77, 49, 31, 38, 38, 77, 46, 34, 41, 45,
31, 77, 42, 27, 45, 45, 35, 41, 40, 45, 77, 44,
31, 27, 30, 79, 77, 77, 77, 77, 77, 77, 77, 77,
23, 34, 35, 29, 34, 77, 51, 31, 46, 77, 45, 47,
44, 48, 35, 48, 31, 64, 77, 45, 46, 27, 39, 42,
31, 30, 77, 41, 40, 77, 46, 34, 31, 45, 31, 77,
38, 35, 32, 31, 38, 31, 45, 45, 77, 46, 34, 35,
40, 33, 45, 64, 79, 77, 77, 77, 77, 77, 77, 77,
77, 20, 34, 31, 77, 34, 27, 40, 30, 77, 46, 34,
27, 46, 77, 39, 41, 29, 37, 31, 30, 77, 46, 34,
31, 39, 64, 77, 27, 40, 30, 77, 46, 34, 31, 77,
34, 31, 27, 44, 46, 77, 46, 34, 27, 46, 77, 32,
31, 30, 66, 79, 77, 77, 77, 77, 77, 77, 77, 77,
1, 40, 30, 77, 41, 40, 77, 46, 34, 31, 77, 42,
31, 30, 31, 45, 46, 27, 38, 64, 77, 46, 34, 31,
45, 31, 77, 49, 41, 44, 30, 45, 77, 27, 42, 42,
31, 27, 44, 65, 79, 77, 77, 77, 77, 77, 77, 77,
77, 13, 51, 77, 40, 27, 39, 31, 77, 35, 45, 77,
15, 52, 51, 39, 27, 40, 30, 35, 27, 45, 64, 77,
11, 35, 40, 33, 77, 41, 32, 77, 11, 35, 40, 33,
45, 66, 79, 77, 77, 77, 77, 77, 77, 77, 77, 12,
41, 41, 37, 77, 41, 40, 77, 39, 51, 77, 23, 41,
44, 37, 45, 64, 77, 51, 31, 77, 13, 35, 33, 34,
46, 51, 64, 77, 27, 40, 30, 77, 30, 31, 45, 42,
27, 35, 44, 67, 79, 77, 77, 77, 77, 77, 77, 77,
77, 14, 41, 46, 34, 35, 40, 33, 77, 28, 31, 45,
35, 30, 31, 77, 44, 31, 39, 27, 35, 40, 45, 63,
77, 18, 41, 47, 40, 30, 77, 46, 34, 31, 77, 30,
31, 29, 27, 51, 79, 77, 77, 77, 77, 77, 77, 77,
77, 15, 32, 77, 46, 34, 27, 46, 77, 29, 41, 38,
41, 45, 45, 27, 38, 77, 23, 44, 31, 29, 37, 64,
77, 28, 41, 47, 40, 30, 38, 31, 45, 45, 77, 27,
40, 30, 77, 28, 27, 44, 31, 79, 77, 77, 77, 77,
77, 77, 77, 77, 20, 34, 31, 77, 38, 41, 40, 31,
77, 27, 40, 30, 77, 38, 31, 48, 31, 38, 77, 45,
27, 40, 30, 45, 77, 45, 46, 44, 31, 46, 29, 34,
77, 32, 27, 44, 77, 27, 49, 27, 51, 79, 77, 77,
77, 77, 77, 77, 77, 77]]),
torch.tensor([[0, 0, 0, 1069, 11, -1, -1, -1, -1]]),
torch.tensor([[0, 0, 0, 1069, 11, -1, -1, -1, -1]]),
]
# fmt: on
self.assertTrue(torch.allclose(tokens[0], EXPECTED_OUTPUT[0]))
self.assertTrue(torch.allclose(tokens[1], EXPECTED_OUTPUT[1]))
self.assertTrue(torch.allclose(tokens[2], EXPECTED_OUTPUT[2]))
utils/check_repo.py
View file @ 61a51f5f
...@@ -51,6 +51,8 @@ IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [ ...@@ -51,6 +51,8 @@ IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
"TableTransformerDecoder", # Building part of bigger (tested) model. "TableTransformerDecoder", # Building part of bigger (tested) model.
"TimeSeriesTransformerEncoder", # Building part of bigger (tested) model. "TimeSeriesTransformerEncoder", # Building part of bigger (tested) model.
"TimeSeriesTransformerDecoder", # Building part of bigger (tested) model. "TimeSeriesTransformerDecoder", # Building part of bigger (tested) model.
"JukeboxVQVAE", # Building part of bigger (tested) model.
"JukeboxPrior", # Building part of bigger (tested) model.
"DeformableDetrEncoder", # Building part of bigger (tested) model. "DeformableDetrEncoder", # Building part of bigger (tested) model.
"DeformableDetrDecoder", # Building part of bigger (tested) model. "DeformableDetrDecoder", # Building part of bigger (tested) model.
"OPTDecoder", # Building part of bigger (tested) model. "OPTDecoder", # Building part of bigger (tested) model.
...@@ -146,6 +148,8 @@ IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [ ...@@ -146,6 +148,8 @@ IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
"CLIPSegTextModel", "CLIPSegTextModel",
"EsmForProteinFolding", "EsmForProteinFolding",
"TimeSeriesTransformerForPrediction", "TimeSeriesTransformerForPrediction",
"JukeboxVQVAE",
"JukeboxPrior",
"PegasusXEncoder", "PegasusXEncoder",
"PegasusXDecoder", "PegasusXDecoder",
"PegasusXDecoderWrapper", "PegasusXDecoderWrapper",
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment