pere commited on
Commit
519e97c
1 Parent(s): 1e886df

Commit from model create scripts

Browse files
.gitattributes CHANGED
@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
28
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
config.gin ADDED
@@ -0,0 +1,147 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __gin__ import dynamic_registration
2
+ import __main__ as train_script
3
+ import seqio
4
+ import t5.data.mixtures
5
+ from t5x import adafactor
6
+ from t5x.examples.t5 import network
7
+ from t5x import gin_utils
8
+ from t5x import models
9
+ from t5x import partitioning
10
+ from t5x import trainer
11
+ from t5x import utils
12
+ import tasks
13
+
14
+ # Macros:
15
+ # ==============================================================================
16
+ BATCH_SIZE = 128
17
+ DROPOUT_RATE = 0.0
18
+ INITIAL_CHECKPOINT_PATH = \
19
+ 'gs://t5-data/pretrained_models/byt5/base/model.ckpt-1000000'
20
+ LABEL_SMOOTHING = 0.0
21
+ LOSS_NORMALIZING_FACTOR = 193536
22
+ MIXTURE_OR_TASK_MODULE = None
23
+ MIXTURE_OR_TASK_NAME = 'byt5_ncc_english_span_corruption_stream'
24
+ MODEL = @models.EncoderDecoderModel()
25
+ MODEL_DIR = \
26
+ 'gs://t5x-training/pretrained_models/norwegian_NCC_plus_English_byt5x_base'
27
+ OPTIMIZER = @adafactor.Adafactor()
28
+ RANDOM_SEED = None
29
+ SHUFFLE_TRAIN_EXAMPLES = True
30
+ TASK_FEATURE_LENGTHS = {'inputs': 512, 'targets': 512}
31
+ TRAIN_STEPS = 1500000
32
+ USE_CACHED_TASKS = True
33
+ USE_HARDWARE_RNG = False
34
+ VOCABULARY = @seqio.ByteVocabulary()
35
+ Z_LOSS = 0.0001
36
+
37
+ # Parameters for adafactor.Adafactor:
38
+ # ==============================================================================
39
+ adafactor.Adafactor.decay_rate = 0.8
40
+ adafactor.Adafactor.logical_factor_rules = \
41
+ @adafactor.standard_logical_factor_rules()
42
+ adafactor.Adafactor.step_offset = 0
43
+
44
+ # Parameters for utils.CheckpointConfig:
45
+ # ==============================================================================
46
+ utils.CheckpointConfig.restore = @utils.RestoreCheckpointConfig()
47
+ utils.CheckpointConfig.save = @utils.SaveCheckpointConfig()
48
+
49
+ # Parameters for utils.create_learning_rate_scheduler:
50
+ # ==============================================================================
51
+ utils.create_learning_rate_scheduler.base_learning_rate = 0.5
52
+ utils.create_learning_rate_scheduler.factors = 'constant * rsqrt_decay'
53
+ utils.create_learning_rate_scheduler.warmup_steps = 10000
54
+
55
+ # Parameters for train/utils.DatasetConfig:
56
+ # ==============================================================================
57
+ train/utils.DatasetConfig.batch_size = %BATCH_SIZE
58
+ train/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
59
+ train/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
60
+ train/utils.DatasetConfig.pack = True
61
+ train/utils.DatasetConfig.seed = None
62
+ train/utils.DatasetConfig.shuffle = %SHUFFLE_TRAIN_EXAMPLES
63
+ train/utils.DatasetConfig.split = 'train'
64
+ train/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
65
+ train/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
66
+
67
+ # Parameters for train_eval/utils.DatasetConfig:
68
+ # ==============================================================================
69
+ train_eval/utils.DatasetConfig.batch_size = %BATCH_SIZE
70
+ train_eval/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
71
+ train_eval/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
72
+ train_eval/utils.DatasetConfig.pack = True
73
+ train_eval/utils.DatasetConfig.seed = 42
74
+ train_eval/utils.DatasetConfig.shuffle = False
75
+ train_eval/utils.DatasetConfig.split = 'validation'
76
+ train_eval/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
77
+ train_eval/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
78
+
79
+ # Parameters for models.EncoderDecoderModel:
80
+ # ==============================================================================
81
+ models.EncoderDecoderModel.input_vocabulary = %VOCABULARY
82
+ models.EncoderDecoderModel.label_smoothing = %LABEL_SMOOTHING
83
+ models.EncoderDecoderModel.loss_normalizing_factor = %LOSS_NORMALIZING_FACTOR
84
+ models.EncoderDecoderModel.module = @network.Transformer()
85
+ models.EncoderDecoderModel.optimizer_def = %OPTIMIZER
86
+ models.EncoderDecoderModel.output_vocabulary = %VOCABULARY
87
+ models.EncoderDecoderModel.z_loss = %Z_LOSS
88
+
89
+ # Parameters for partitioning.PjitPartitioner:
90
+ # ==============================================================================
91
+ partitioning.PjitPartitioner.logical_axis_rules = \
92
+ @partitioning.standard_logical_axis_rules()
93
+ partitioning.PjitPartitioner.model_parallel_submesh = None
94
+ partitioning.PjitPartitioner.num_partitions = 1
95
+
96
+ # Parameters for utils.RestoreCheckpointConfig:
97
+ # ==============================================================================
98
+ utils.RestoreCheckpointConfig.dtype = 'float32'
99
+ utils.RestoreCheckpointConfig.mode = 'specific'
100
+ utils.RestoreCheckpointConfig.path = %INITIAL_CHECKPOINT_PATH
101
+
102
+ # Parameters for utils.SaveCheckpointConfig:
103
+ # ==============================================================================
104
+ utils.SaveCheckpointConfig.dtype = 'float32'
105
+ utils.SaveCheckpointConfig.keep = None
106
+ utils.SaveCheckpointConfig.period = 1000
107
+ utils.SaveCheckpointConfig.save_dataset = False
108
+
109
+ # Parameters for network.T5Config:
110
+ # ==============================================================================
111
+ network.T5Config.dropout_rate = %DROPOUT_RATE
112
+ network.T5Config.dtype = 'bfloat16'
113
+ network.T5Config.emb_dim = 1536
114
+ network.T5Config.head_dim = 64
115
+ network.T5Config.logits_via_embedding = False
116
+ network.T5Config.mlp_activations = ('gelu', 'linear')
117
+ network.T5Config.mlp_dim = 3968
118
+ network.T5Config.num_decoder_layers = 6
119
+ network.T5Config.num_encoder_layers = 18
120
+ network.T5Config.num_heads = 12
121
+ network.T5Config.vocab_size = 384
122
+
123
+ # Parameters for train_script.train:
124
+ # ==============================================================================
125
+ train_script.train.checkpoint_cfg = @utils.CheckpointConfig()
126
+ train_script.train.eval_period = 1000
127
+ train_script.train.eval_steps = 20
128
+ train_script.train.infer_eval_dataset_cfg = None
129
+ train_script.train.model = %MODEL
130
+ train_script.train.model_dir = %MODEL_DIR
131
+ train_script.train.partitioner = @partitioning.PjitPartitioner()
132
+ train_script.train.random_seed = %RANDOM_SEED
133
+ train_script.train.summarize_config_fn = @gin_utils.summarize_gin_config
134
+ train_script.train.total_steps = %TRAIN_STEPS
135
+ train_script.train.train_dataset_cfg = @train/utils.DatasetConfig()
136
+ train_script.train.train_eval_dataset_cfg = @train_eval/utils.DatasetConfig()
137
+ train_script.train.trainer_cls = @trainer.Trainer
138
+ train_script.train.use_hardware_rng = %USE_HARDWARE_RNG
139
+
140
+ # Parameters for trainer.Trainer:
141
+ # ==============================================================================
142
+ trainer.Trainer.learning_rate_fn = @utils.create_learning_rate_scheduler()
143
+ trainer.Trainer.num_microbatches = None
144
+
145
+ # Parameters for network.Transformer:
146
+ # ==============================================================================
147
+ network.Transformer.config = @network.T5Config()
config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/home/perk/models/byt5_base_NCC",
3
+ "architectures": [
4
+ "T5ForConditionalGeneration"
5
+ ],
6
+ "d_ff": 3968,
7
+ "d_kv": 64,
8
+ "d_model": 1536,
9
+ "decoder_start_token_id": 0,
10
+ "dropout_rate": 0.1,
11
+ "eos_token_id": 1,
12
+ "feed_forward_proj": "gated-gelu",
13
+ "gradient_checkpointing": false,
14
+ "initializer_factor": 1.0,
15
+ "is_encoder_decoder": true,
16
+ "layer_norm_epsilon": 1e-06,
17
+ "model_type": "t5",
18
+ "num_decoder_layers": 6,
19
+ "num_heads": 12,
20
+ "num_layers": 18,
21
+ "output_past": true,
22
+ "pad_token_id": 0,
23
+ "relative_attention_max_distance": 128,
24
+ "relative_attention_num_buckets": 32,
25
+ "tie_word_embeddings": false,
26
+ "tokenizer_class": "ByT5Tokenizer",
27
+ "torch_dtype": "float32",
28
+ "transformers_version": "4.19.2",
29
+ "use_cache": true,
30
+ "vocab_size": 384
31
+ }
flax_model.msgpack ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:055bba8e7d7e936a3e563a9ddabe78e4c236e13dc127c1b9c55deeb57c00cebe
3
+ size 2326623947
model-info.txt ADDED
The diff for this file is too large to render. See raw diff
 
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90728b46881d00abdea86a164bf4c959a88b8f90462a0810bed6e20567af9fd3
3
+ size 2326694409
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}
spiece.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef78f86560d809067d12bac6c09f19a462cb3af3f54d2b8acbba26e1433125d6
3
+ size 4309802
tf_model.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:968e4dfd9da58aaad02eca7656a472c46f981be3d39f1fb29cc93e7eae8a4b92
3
+ size 2327127104
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "extra_ids": 0, "additional_special_tokens": null, "special_tokens_map_file": "/home/patrick/.cache/torch/transformers/685ac0ca8568ec593a48b61b0a3c272beee9bc194a3c7241d15dcadb5f875e53.f76030f3ec1b96a8199b2593390c610e76ca8028ef3d24680000619ffb646276", "tokenizer_file": null, "name_or_path": "google/mt5-small"}
train/events.out.tfevents.1651948511.t1v-n-690df9bb-w-0.2369704.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e1612e32da33d510bbcbb0cc0e918d3c00d39e33b5906d8c2a67709b2934dbc
3
+ size 6146
train/events.out.tfevents.1651948631.t1v-n-690df9bb-w-0.2371380.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f58e28c4d7cf0b28ed2ff2a76b7af5b6a5885ad43fd8af6ac4e32c67882a1531
3
+ size 84976
train/events.out.tfevents.1652004005.t1v-n-690df9bb-w-0.2538803.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2da4f6a4b431fb45ddcc99d5387e1c050851a153a9bce66c207d774bbec254db
3
+ size 494136
train/events.out.tfevents.1652278024.t1v-n-690df9bb-w-0.23378.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bc69901de0aadfc39c509ee176bfc8f8bd9ada5857e1b33ca67b5e2859356dd
3
+ size 58728
train/events.out.tfevents.1652333049.t1v-n-eaf6c591-w-0.10368.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30d1798b6d7d4fb0b865c719567a5fab13c4f714d7c8765d2727629a625ed295
3
+ size 159088
training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1651948512.t1v-n-690df9bb-w-0.2369704.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:14363ddea0c625ec6f45939c124e6fdc0050a43564f5dd2423bd83aa5878e5cb
3
+ size 40
training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1651948632.t1v-n-690df9bb-w-0.2371380.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02f4ca66df4cb44578b0af1ffcd1168ac8b314753e57ab6980ed9c34e5b6ad2d
3
+ size 66681
training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652004005.t1v-n-690df9bb-w-0.2538803.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24325cb7f945e38d476f50c340ba03f21458fdbcc3ed607899902df07c0d02da
3
+ size 412506
training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652278024.t1v-n-690df9bb-w-0.23378.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57fe574e64c5e25055cbee94289c39c743ddca9c6f6a65dd47212c478dcaa00f
3
+ size 44496
training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652333050.t1v-n-eaf6c591-w-0.10368.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccd5ee8d99d32ae03f37b55c3ce500a0f79152ae2847b283e26c5582f9a55345
3
+ size 129321