Spaces:
Running
Running
fix: adjust training script + dataloader
Browse files
dalle_mini/data.py
CHANGED
@@ -15,12 +15,10 @@ class Dataset:
|
|
15 |
dataset_repo_or_path: str
|
16 |
train_file: str = None
|
17 |
validation_file: str = None
|
18 |
-
dataset_type: str = "dataset"
|
19 |
streaming: bool = True
|
20 |
use_auth_token: bool = False
|
21 |
text_column: str = "caption"
|
22 |
encoding_column: str = "encoding"
|
23 |
-
max_source_length: int = 128
|
24 |
max_train_samples: int = None
|
25 |
max_eval_samples: int = None
|
26 |
preprocessing_num_workers: int = None
|
@@ -70,7 +68,7 @@ class Dataset:
|
|
70 |
else self.eval_dataset.select(range(self.max_eval_samples))
|
71 |
)
|
72 |
|
73 |
-
def preprocess(self, tokenizer, decoder_start_token_id, normalize_text):
|
74 |
if self.streaming:
|
75 |
# we need to shuffle early in streaming mode
|
76 |
if hasattr(self, "train_dataset"):
|
@@ -112,7 +110,7 @@ class Dataset:
|
|
112 |
tokenizer=tokenizer,
|
113 |
text_column=self.text_column,
|
114 |
encoding_column=self.encoding_column,
|
115 |
-
|
116 |
decoder_start_token_id=decoder_start_token_id,
|
117 |
)
|
118 |
for ds in ["train_dataset", "eval_dataset"]:
|
@@ -232,14 +230,14 @@ def preprocess_function(
|
|
232 |
tokenizer,
|
233 |
text_column,
|
234 |
encoding_column,
|
235 |
-
|
236 |
decoder_start_token_id,
|
237 |
):
|
238 |
inputs = examples[text_column]
|
239 |
# Setting padding="max_length" as we need fixed length inputs for jitted functions
|
240 |
model_inputs = tokenizer(
|
241 |
inputs,
|
242 |
-
max_length=
|
243 |
padding="max_length",
|
244 |
truncation=True,
|
245 |
return_tensors="np",
|
|
|
15 |
dataset_repo_or_path: str
|
16 |
train_file: str = None
|
17 |
validation_file: str = None
|
|
|
18 |
streaming: bool = True
|
19 |
use_auth_token: bool = False
|
20 |
text_column: str = "caption"
|
21 |
encoding_column: str = "encoding"
|
|
|
22 |
max_train_samples: int = None
|
23 |
max_eval_samples: int = None
|
24 |
preprocessing_num_workers: int = None
|
|
|
68 |
else self.eval_dataset.select(range(self.max_eval_samples))
|
69 |
)
|
70 |
|
71 |
+
def preprocess(self, tokenizer, decoder_start_token_id, normalize_text, max_length):
|
72 |
if self.streaming:
|
73 |
# we need to shuffle early in streaming mode
|
74 |
if hasattr(self, "train_dataset"):
|
|
|
110 |
tokenizer=tokenizer,
|
111 |
text_column=self.text_column,
|
112 |
encoding_column=self.encoding_column,
|
113 |
+
max_length=max_length,
|
114 |
decoder_start_token_id=decoder_start_token_id,
|
115 |
)
|
116 |
for ds in ["train_dataset", "eval_dataset"]:
|
|
|
230 |
tokenizer,
|
231 |
text_column,
|
232 |
encoding_column,
|
233 |
+
max_length,
|
234 |
decoder_start_token_id,
|
235 |
):
|
236 |
inputs = examples[text_column]
|
237 |
# Setting padding="max_length" as we need fixed length inputs for jitted functions
|
238 |
model_inputs = tokenizer(
|
239 |
inputs,
|
240 |
+
max_length=max_length,
|
241 |
padding="max_length",
|
242 |
truncation=True,
|
243 |
return_tensors="np",
|
dalle_mini/model.py
DELETED
@@ -1,64 +0,0 @@
|
|
1 |
-
import flax.linen as nn
|
2 |
-
import jax
|
3 |
-
from transformers import BartConfig
|
4 |
-
from transformers.models.bart.modeling_flax_bart import (
|
5 |
-
FlaxBartDecoder,
|
6 |
-
FlaxBartEncoder,
|
7 |
-
FlaxBartForConditionalGeneration,
|
8 |
-
FlaxBartForConditionalGenerationModule,
|
9 |
-
FlaxBartModule,
|
10 |
-
)
|
11 |
-
|
12 |
-
|
13 |
-
class CustomFlaxBartModule(FlaxBartModule):
|
14 |
-
def setup(self):
|
15 |
-
# we keep shared to easily load pre-trained weights
|
16 |
-
self.shared = nn.Embed(
|
17 |
-
self.config.vocab_size,
|
18 |
-
self.config.d_model,
|
19 |
-
embedding_init=jax.nn.initializers.normal(self.config.init_std),
|
20 |
-
)
|
21 |
-
# a separate embedding is used for the decoder
|
22 |
-
self.decoder_embed = nn.Embed(
|
23 |
-
self.config.image_vocab_size + 1,
|
24 |
-
self.config.d_model,
|
25 |
-
embedding_init=jax.nn.initializers.normal(self.config.init_std),
|
26 |
-
)
|
27 |
-
self.encoder = FlaxBartEncoder(
|
28 |
-
self.config, dtype=self.dtype, embed_tokens=self.shared
|
29 |
-
)
|
30 |
-
|
31 |
-
# the decoder has a different config
|
32 |
-
# TODO: should not be needed once we have custom config/module
|
33 |
-
decoder_config = BartConfig(self.config.to_dict())
|
34 |
-
decoder_config.max_position_embeddings = (
|
35 |
-
self.config.image_length + 1 # image tokens + BOS
|
36 |
-
)
|
37 |
-
decoder_config.vocab_size = self.config.image_vocab_size + 1
|
38 |
-
self.decoder = FlaxBartDecoder(
|
39 |
-
decoder_config, dtype=self.dtype, embed_tokens=self.decoder_embed
|
40 |
-
)
|
41 |
-
|
42 |
-
|
43 |
-
class CustomFlaxBartForConditionalGenerationModule(
|
44 |
-
FlaxBartForConditionalGenerationModule
|
45 |
-
):
|
46 |
-
def setup(self):
|
47 |
-
# set default config
|
48 |
-
self.config.normalize_text = getattr(self.config, "normalize_text", False)
|
49 |
-
self.config.image_length = getattr(self.config, "image_length", 256)
|
50 |
-
self.config.image_vocab_size = getattr(self.config, "image_vocab_size", 16384)
|
51 |
-
|
52 |
-
self.model = CustomFlaxBartModule(config=self.config, dtype=self.dtype)
|
53 |
-
self.lm_head = nn.Dense(
|
54 |
-
self.config.image_vocab_size + 1, # encoded image token space + 1 for bos
|
55 |
-
use_bias=False,
|
56 |
-
kernel_init=jax.nn.initializers.normal(self.config.init_std),
|
57 |
-
)
|
58 |
-
self.final_logits_bias = self.param(
|
59 |
-
"final_logits_bias", self.bias_init, (1, self.config.image_vocab_size + 1)
|
60 |
-
)
|
61 |
-
|
62 |
-
|
63 |
-
class CustomFlaxBartForConditionalGeneration(FlaxBartForConditionalGeneration):
|
64 |
-
module_class = CustomFlaxBartForConditionalGenerationModule
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dalle_mini/model/__init__.py
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
from .configuration import DalleBartConfig
|
2 |
+
from .modeling import DalleBartForConditionalGeneration
|
dalle_mini/{configuration_bart.py → model/configuration.py}
RENAMED
@@ -12,7 +12,7 @@
|
|
12 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
-
"""
|
16 |
import warnings
|
17 |
|
18 |
from transformers.configuration_utils import PretrainedConfig
|
@@ -123,7 +123,7 @@ class DalleBartConfig(PretrainedConfig):
|
|
123 |
):
|
124 |
self.normalize_text = normalize_text
|
125 |
self.encoder_vocab_size = encoder_vocab_size
|
126 |
-
self.
|
127 |
self.image_length = image_length
|
128 |
self.max_text_length = max_text_length
|
129 |
self.d_model = d_model
|
@@ -145,17 +145,21 @@ class DalleBartConfig(PretrainedConfig):
|
|
145 |
self.num_hidden_layers = encoder_layers
|
146 |
self.gradient_checkpointing = gradient_checkpointing
|
147 |
self.scale_embedding = scale_embedding # scale factor will be sqrt(d_model) if True
|
148 |
-
self.decoder_start_token_id = image_vocab_size
|
149 |
self.min_length = image_length + 1
|
150 |
self.max_length = image_length + 1
|
151 |
|
|
|
|
|
|
|
|
|
152 |
super().__init__(
|
153 |
num_labels=num_labels,
|
154 |
pad_token_id=image_vocab_size + 1, # needed to avoid errors during generation (converted to jnp.array)
|
155 |
bos_token_id=image_vocab_size + 1, # set to unreachable values
|
156 |
eos_token_id=image_vocab_size + 1,
|
157 |
is_encoder_decoder=is_encoder_decoder,
|
158 |
-
decoder_start_token_id=decoder_start_token_id,
|
159 |
forced_eos_token_id=forced_eos_token_id,
|
160 |
tie_word_embeddings=tie_word_embeddings,
|
161 |
**kwargs,
|
|
|
12 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
+
""" DalleBart model configuration """
|
16 |
import warnings
|
17 |
|
18 |
from transformers.configuration_utils import PretrainedConfig
|
|
|
123 |
):
|
124 |
self.normalize_text = normalize_text
|
125 |
self.encoder_vocab_size = encoder_vocab_size
|
126 |
+
self.image_vocab_size = image_vocab_size
|
127 |
self.image_length = image_length
|
128 |
self.max_text_length = max_text_length
|
129 |
self.d_model = d_model
|
|
|
145 |
self.num_hidden_layers = encoder_layers
|
146 |
self.gradient_checkpointing = gradient_checkpointing
|
147 |
self.scale_embedding = scale_embedding # scale factor will be sqrt(d_model) if True
|
148 |
+
self.decoder_start_token_id = image_vocab_size # BOS appended to vocab
|
149 |
self.min_length = image_length + 1
|
150 |
self.max_length = image_length + 1
|
151 |
|
152 |
+
# remove keys we are about to set to prevent errors
|
153 |
+
for k in ['bos_token_id', 'eos_token_id', 'pad_token_id', 'decoder_start_token_id', 'forced_eos_token_id']:
|
154 |
+
kwargs.pop(k, None)
|
155 |
+
|
156 |
super().__init__(
|
157 |
num_labels=num_labels,
|
158 |
pad_token_id=image_vocab_size + 1, # needed to avoid errors during generation (converted to jnp.array)
|
159 |
bos_token_id=image_vocab_size + 1, # set to unreachable values
|
160 |
eos_token_id=image_vocab_size + 1,
|
161 |
is_encoder_decoder=is_encoder_decoder,
|
162 |
+
decoder_start_token_id=self.decoder_start_token_id,
|
163 |
forced_eos_token_id=forced_eos_token_id,
|
164 |
tie_word_embeddings=tie_word_embeddings,
|
165 |
**kwargs,
|
dalle_mini/{modeling_bart_flax.py → model/modeling.py}
RENAMED
@@ -45,7 +45,7 @@ from transformers.modeling_flax_utils import (
|
|
45 |
from transformers.utils import logging
|
46 |
|
47 |
|
48 |
-
from .
|
49 |
|
50 |
|
51 |
logger = logging.get_logger(__name__)
|
@@ -64,7 +64,7 @@ def shift_tokens_right(input_ids: np.array, pad_token_id: int, decoder_start_tok
|
|
64 |
|
65 |
|
66 |
class FlaxBartAttention(nn.Module):
|
67 |
-
config:
|
68 |
embed_dim: int
|
69 |
num_heads: int
|
70 |
dropout: float = 0.0
|
@@ -93,7 +93,7 @@ class FlaxBartAttention(nn.Module):
|
|
93 |
|
94 |
if self.causal:
|
95 |
self.causal_mask = make_causal_mask(
|
96 |
-
jnp.ones((1, embed_dim), dtype="bool"), dtype="bool"
|
97 |
)
|
98 |
|
99 |
def _split_heads(self, hidden_states):
|
@@ -224,7 +224,7 @@ class FlaxBartAttention(nn.Module):
|
|
224 |
|
225 |
|
226 |
class FlaxBartEncoderLayer(nn.Module):
|
227 |
-
config:
|
228 |
dtype: jnp.dtype = jnp.float32
|
229 |
|
230 |
def setup(self) -> None:
|
@@ -279,7 +279,7 @@ class FlaxBartEncoderLayer(nn.Module):
|
|
279 |
|
280 |
|
281 |
class FlaxBartEncoderLayerCollection(nn.Module):
|
282 |
-
config:
|
283 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
284 |
|
285 |
def setup(self):
|
@@ -306,7 +306,7 @@ class FlaxBartEncoderLayerCollection(nn.Module):
|
|
306 |
|
307 |
|
308 |
class FlaxBartDecoderLayer(nn.Module):
|
309 |
-
config:
|
310 |
dtype: jnp.dtype = jnp.float32
|
311 |
|
312 |
def setup(self) -> None:
|
@@ -390,7 +390,7 @@ class FlaxBartDecoderLayer(nn.Module):
|
|
390 |
|
391 |
|
392 |
class FlaxBartDecoderLayerCollection(nn.Module):
|
393 |
-
config:
|
394 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
395 |
|
396 |
def setup(self):
|
@@ -422,8 +422,8 @@ class FlaxBartDecoderLayerCollection(nn.Module):
|
|
422 |
return FlaxBaseModelOutputWithPastAndCrossAttentions(last_hidden_state=hidden_states)
|
423 |
|
424 |
|
425 |
-
class
|
426 |
-
config:
|
427 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
428 |
|
429 |
def setup(self):
|
@@ -479,8 +479,8 @@ class FlaxBartEncoder(nn.Module):
|
|
479 |
)
|
480 |
|
481 |
|
482 |
-
class
|
483 |
-
config:
|
484 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
485 |
|
486 |
def setup(self):
|
@@ -550,13 +550,13 @@ class FlaxBartDecoder(nn.Module):
|
|
550 |
)
|
551 |
|
552 |
|
553 |
-
class
|
554 |
-
config:
|
555 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
556 |
|
557 |
def setup(self):
|
558 |
-
self.encoder =
|
559 |
-
self.decoder =
|
560 |
|
561 |
def _get_encoder_module(self):
|
562 |
return self.encoder
|
@@ -605,14 +605,14 @@ class FlaxBartModule(nn.Module):
|
|
605 |
)
|
606 |
|
607 |
|
608 |
-
class
|
609 |
-
config_class =
|
610 |
-
base_model_prefix: str = "
|
611 |
module_class: nn.Module = None
|
612 |
|
613 |
def __init__(
|
614 |
self,
|
615 |
-
config:
|
616 |
input_shape: Tuple[int] = (1, 1),
|
617 |
seed: int = 0,
|
618 |
dtype: jnp.dtype = jnp.float32,
|
@@ -792,13 +792,13 @@ class FlaxBartPreTrainedModel(FlaxPreTrainedModel):
|
|
792 |
)
|
793 |
|
794 |
|
795 |
-
class
|
796 |
-
config:
|
797 |
dtype: jnp.dtype = jnp.float32
|
798 |
bias_init: Callable[..., jnp.ndarray] = jax.nn.initializers.zeros
|
799 |
|
800 |
def setup(self):
|
801 |
-
self.model =
|
802 |
self.lm_head = nn.Dense(
|
803 |
self.config.image_vocab_size + 1, # image vocab size + 1 for BOS
|
804 |
use_bias=False,
|
@@ -854,8 +854,8 @@ class FlaxBartForConditionalGenerationModule(nn.Module):
|
|
854 |
)
|
855 |
|
856 |
|
857 |
-
class
|
858 |
-
module_class =
|
859 |
dtype: jnp.dtype = jnp.float32
|
860 |
|
861 |
def decode(
|
|
|
45 |
from transformers.utils import logging
|
46 |
|
47 |
|
48 |
+
from .configuration import DalleBartConfig
|
49 |
|
50 |
|
51 |
logger = logging.get_logger(__name__)
|
|
|
64 |
|
65 |
|
66 |
class FlaxBartAttention(nn.Module):
|
67 |
+
config: DalleBartConfig
|
68 |
embed_dim: int
|
69 |
num_heads: int
|
70 |
dropout: float = 0.0
|
|
|
93 |
|
94 |
if self.causal:
|
95 |
self.causal_mask = make_causal_mask(
|
96 |
+
jnp.ones((1, self.embed_dim), dtype="bool"), dtype="bool"
|
97 |
)
|
98 |
|
99 |
def _split_heads(self, hidden_states):
|
|
|
224 |
|
225 |
|
226 |
class FlaxBartEncoderLayer(nn.Module):
|
227 |
+
config: DalleBartConfig
|
228 |
dtype: jnp.dtype = jnp.float32
|
229 |
|
230 |
def setup(self) -> None:
|
|
|
279 |
|
280 |
|
281 |
class FlaxBartEncoderLayerCollection(nn.Module):
|
282 |
+
config: DalleBartConfig
|
283 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
284 |
|
285 |
def setup(self):
|
|
|
306 |
|
307 |
|
308 |
class FlaxBartDecoderLayer(nn.Module):
|
309 |
+
config: DalleBartConfig
|
310 |
dtype: jnp.dtype = jnp.float32
|
311 |
|
312 |
def setup(self) -> None:
|
|
|
390 |
|
391 |
|
392 |
class FlaxBartDecoderLayerCollection(nn.Module):
|
393 |
+
config: DalleBartConfig
|
394 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
395 |
|
396 |
def setup(self):
|
|
|
422 |
return FlaxBaseModelOutputWithPastAndCrossAttentions(last_hidden_state=hidden_states)
|
423 |
|
424 |
|
425 |
+
class DalleBartEncoder(nn.Module):
|
426 |
+
config: DalleBartConfig
|
427 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
428 |
|
429 |
def setup(self):
|
|
|
479 |
)
|
480 |
|
481 |
|
482 |
+
class DalleBartDecoder(nn.Module):
|
483 |
+
config: DalleBartConfig
|
484 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
485 |
|
486 |
def setup(self):
|
|
|
550 |
)
|
551 |
|
552 |
|
553 |
+
class DalleBartModule(nn.Module):
|
554 |
+
config: DalleBartConfig
|
555 |
dtype: jnp.dtype = jnp.float32 # the dtype of the computation
|
556 |
|
557 |
def setup(self):
|
558 |
+
self.encoder = DalleBartEncoder(self.config, dtype=self.dtype)
|
559 |
+
self.decoder = DalleBartDecoder(self.config, dtype=self.dtype)
|
560 |
|
561 |
def _get_encoder_module(self):
|
562 |
return self.encoder
|
|
|
605 |
)
|
606 |
|
607 |
|
608 |
+
class DalleBartPreTrainedModel(FlaxPreTrainedModel):
|
609 |
+
config_class = DalleBartConfig
|
610 |
+
base_model_prefix: str = "dallebart"
|
611 |
module_class: nn.Module = None
|
612 |
|
613 |
def __init__(
|
614 |
self,
|
615 |
+
config: DalleBartConfig,
|
616 |
input_shape: Tuple[int] = (1, 1),
|
617 |
seed: int = 0,
|
618 |
dtype: jnp.dtype = jnp.float32,
|
|
|
792 |
)
|
793 |
|
794 |
|
795 |
+
class DalleBartForConditionalGenerationModule(nn.Module):
|
796 |
+
config: DalleBartConfig
|
797 |
dtype: jnp.dtype = jnp.float32
|
798 |
bias_init: Callable[..., jnp.ndarray] = jax.nn.initializers.zeros
|
799 |
|
800 |
def setup(self):
|
801 |
+
self.model = DalleBartModule(config=self.config, dtype=self.dtype)
|
802 |
self.lm_head = nn.Dense(
|
803 |
self.config.image_vocab_size + 1, # image vocab size + 1 for BOS
|
804 |
use_bias=False,
|
|
|
854 |
)
|
855 |
|
856 |
|
857 |
+
class DalleBartForConditionalGeneration(DalleBartPreTrainedModel):
|
858 |
+
module_class = DalleBartForConditionalGenerationModule
|
859 |
dtype: jnp.dtype = jnp.float32
|
860 |
|
861 |
def decode(
|
dalle_mini/{partitions.py → model/partitions.py}
RENAMED
@@ -5,7 +5,7 @@ from flax.traverse_util import flatten_dict, unflatten_dict
|
|
5 |
from jax.experimental import PartitionSpec as P
|
6 |
|
7 |
|
8 |
-
# utils adapted from https://
|
9 |
# Sentinels
|
10 |
_unmatched = object()
|
11 |
|
|
|
5 |
from jax.experimental import PartitionSpec as P
|
6 |
|
7 |
|
8 |
+
# utils adapted from https://github.com/google-research/google-research/blob/master/flax_models/t5x/partitions.py
|
9 |
# Sentinels
|
10 |
_unmatched = object()
|
11 |
|
tools/train/train.py
CHANGED
@@ -44,7 +44,7 @@ from transformers import AutoTokenizer, HfArgumentParser
|
|
44 |
from transformers.models.bart.modeling_flax_bart import BartConfig
|
45 |
|
46 |
from dalle_mini.data import Dataset
|
47 |
-
from dalle_mini.model import
|
48 |
|
49 |
logger = logging.getLogger(__name__)
|
50 |
|
@@ -68,26 +68,12 @@ class ModelArguments:
|
|
68 |
"help": "Pretrained config name or path if not the same as model_name"
|
69 |
},
|
70 |
)
|
71 |
-
image_vocab_size: Optional[int] = field(
|
72 |
-
default=None,
|
73 |
-
metadata={"help": "Vocab size of image encoder"},
|
74 |
-
)
|
75 |
-
image_length: Optional[int] = field(
|
76 |
-
default=None,
|
77 |
-
metadata={"help": "Number of tokens per image"},
|
78 |
-
)
|
79 |
tokenizer_name: Optional[str] = field(
|
80 |
default=None,
|
81 |
metadata={
|
82 |
"help": "Pretrained tokenizer name or path if not the same as model_name_or_path"
|
83 |
},
|
84 |
)
|
85 |
-
normalize_text: Optional[bool] = field(
|
86 |
-
default=None,
|
87 |
-
metadata={
|
88 |
-
"help": "Whether to normalize text or not. By default, we refer to base model or don't normalize for new models."
|
89 |
-
},
|
90 |
-
)
|
91 |
dtype: Optional[str] = field(
|
92 |
default="float32",
|
93 |
metadata={
|
@@ -126,10 +112,6 @@ class DataTrainingArguments:
|
|
126 |
default=None,
|
127 |
metadata={"help": "An optional input evaluation data file (glob acceptable)."},
|
128 |
)
|
129 |
-
dataset_type: str = field(
|
130 |
-
default="datasets",
|
131 |
-
metadata={"help": "Either 🤗 'dataset' (default) or 'webdataset'."},
|
132 |
-
)
|
133 |
# data loading should not be a bottleneck so we use "streaming" mode by default
|
134 |
streaming: bool = field(
|
135 |
default=True,
|
@@ -141,13 +123,6 @@ class DataTrainingArguments:
|
|
141 |
"help": "Whether to use the authentication token for private datasets."
|
142 |
},
|
143 |
)
|
144 |
-
max_source_length: Optional[int] = field(
|
145 |
-
default=128,
|
146 |
-
metadata={
|
147 |
-
"help": "The maximum total input sequence length after tokenization. Sequences longer "
|
148 |
-
"than this will be truncated, sequences shorter will be padded."
|
149 |
-
},
|
150 |
-
)
|
151 |
max_train_samples: Optional[int] = field(
|
152 |
default=None,
|
153 |
metadata={
|
@@ -436,47 +411,14 @@ def main():
|
|
436 |
|
437 |
else:
|
438 |
# Set up our new model config
|
439 |
-
# TODO: simplify with custom config class
|
440 |
if model_args.config_name:
|
441 |
-
config =
|
442 |
-
else:
|
443 |
-
config = BartConfig.from_pretrained(model_args.model_name_or_path)
|
444 |
-
if model_args.image_vocab_size:
|
445 |
-
config.image_vocab_size = model_args.image_vocab_size
|
446 |
-
assert (
|
447 |
-
getattr(config, "image_vocab_size") is not None
|
448 |
-
), "image_vocab_size must be specified when not present in base model/config"
|
449 |
-
if model_args.image_length:
|
450 |
-
config.image_length = model_args.image_length
|
451 |
-
assert (
|
452 |
-
getattr(config, "image_length") is not None
|
453 |
-
), "image_length must be specified when not present in base model/config"
|
454 |
-
# we append decoder bos to image vocab
|
455 |
-
config.decoder_start_token_id = config.image_vocab_size
|
456 |
-
# ensure we don't generate bos (in addition to decoder start token)
|
457 |
-
config.force_bos_token_to_be_generated = False
|
458 |
-
config.forced_bos_token_id = None # we don't need this token
|
459 |
-
config.forced_eos_token_id = None # we don't need this token
|
460 |
-
|
461 |
-
config.tie_word_embeddings = False
|
462 |
-
config.min_length = config.image_length + 1
|
463 |
-
config.max_length = config.image_length + 1
|
464 |
-
|
465 |
-
# below tokens need to be set to avoid error during generation (converted to jnp.array)
|
466 |
-
# they are not expected to be used and are set to unreachable token id
|
467 |
-
config.bos_token_id = config.image_vocab_size + 1
|
468 |
-
config.pos_token_id = config.image_vocab_size + 1
|
469 |
-
config.eos_token_id = config.image_vocab_size + 1
|
470 |
-
|
471 |
-
# save whether we normalize the text
|
472 |
-
if model_args.normalize_text is not None:
|
473 |
-
config.normalize_text = model_args.normalize_text
|
474 |
else:
|
475 |
-
config
|
476 |
|
477 |
# Load or create new model
|
478 |
if model_args.model_name_or_path:
|
479 |
-
model =
|
480 |
model_args.model_name_or_path,
|
481 |
config=config,
|
482 |
seed=training_args.seed_model,
|
@@ -485,7 +427,7 @@ def main():
|
|
485 |
# avoid OOM on TPU: see https://github.com/google/flax/issues/1658
|
486 |
print(model.params)
|
487 |
else:
|
488 |
-
model =
|
489 |
config,
|
490 |
seed=training_args.seed_model,
|
491 |
dtype=getattr(jnp, model_args.dtype),
|
@@ -512,6 +454,7 @@ def main():
|
|
512 |
tokenizer=tokenizer,
|
513 |
decoder_start_token_id=model.config.decoder_start_token_id,
|
514 |
normalize_text=model.config.normalize_text,
|
|
|
515 |
)
|
516 |
|
517 |
# Initialize our training
|
|
|
44 |
from transformers.models.bart.modeling_flax_bart import BartConfig
|
45 |
|
46 |
from dalle_mini.data import Dataset
|
47 |
+
from dalle_mini.model import DalleBartConfig, DalleBartForConditionalGeneration
|
48 |
|
49 |
logger = logging.getLogger(__name__)
|
50 |
|
|
|
68 |
"help": "Pretrained config name or path if not the same as model_name"
|
69 |
},
|
70 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
tokenizer_name: Optional[str] = field(
|
72 |
default=None,
|
73 |
metadata={
|
74 |
"help": "Pretrained tokenizer name or path if not the same as model_name_or_path"
|
75 |
},
|
76 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
dtype: Optional[str] = field(
|
78 |
default="float32",
|
79 |
metadata={
|
|
|
112 |
default=None,
|
113 |
metadata={"help": "An optional input evaluation data file (glob acceptable)."},
|
114 |
)
|
|
|
|
|
|
|
|
|
115 |
# data loading should not be a bottleneck so we use "streaming" mode by default
|
116 |
streaming: bool = field(
|
117 |
default=True,
|
|
|
123 |
"help": "Whether to use the authentication token for private datasets."
|
124 |
},
|
125 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
max_train_samples: Optional[int] = field(
|
127 |
default=None,
|
128 |
metadata={
|
|
|
411 |
|
412 |
else:
|
413 |
# Set up our new model config
|
|
|
414 |
if model_args.config_name:
|
415 |
+
config = DalleBartConfig.from_pretrained(model_args.config_name)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
416 |
else:
|
417 |
+
config = DalleBartConfig.from_pretrained(model_args.model_name_or_path)
|
418 |
|
419 |
# Load or create new model
|
420 |
if model_args.model_name_or_path:
|
421 |
+
model = DalleBartForConditionalGeneration.from_pretrained(
|
422 |
model_args.model_name_or_path,
|
423 |
config=config,
|
424 |
seed=training_args.seed_model,
|
|
|
427 |
# avoid OOM on TPU: see https://github.com/google/flax/issues/1658
|
428 |
print(model.params)
|
429 |
else:
|
430 |
+
model = DalleBartForConditionalGeneration(
|
431 |
config,
|
432 |
seed=training_args.seed_model,
|
433 |
dtype=getattr(jnp, model_args.dtype),
|
|
|
454 |
tokenizer=tokenizer,
|
455 |
decoder_start_token_id=model.config.decoder_start_token_id,
|
456 |
normalize_text=model.config.normalize_text,
|
457 |
+
max_length=model.config.max_text_length,
|
458 |
)
|
459 |
|
460 |
# Initialize our training
|