pere commited on
Commit
49e61c5
1 Parent(s): ced2459

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 +1,178 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __gin__ import dynamic_registration
2
+ import __main__ as train_script
3
+ import seqio
4
+ from t5.data import 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 = 0.1
18
+ EVAL_STEPS = 20
19
+ EVALUATOR_NUM_EXAMPLES = None
20
+ EVALUATOR_USE_MEMORY_CACHE = True
21
+ INITIAL_CHECKPOINT_PATH = \
22
+ 'gs://north-t5x/pretrained_models/base/norwegian_NCC_plus_English_pluss200k_balanced_bokmaal_nynorsk_t5x_base/checkpoint_1700000'
23
+ JSON_WRITE_N_RESULTS = None
24
+ LABEL_SMOOTHING = 0.0
25
+ LOSS_NORMALIZING_FACTOR = None
26
+ MIXTURE_OR_TASK_MODULE = None
27
+ MIXTURE_OR_TASK_NAME = 'deuncaser'
28
+ MODEL = @models.EncoderDecoderModel()
29
+ MODEL_DIR = 'gs://north-t5x/finetuned/deuncaser/deuncaser_base_v1'
30
+ OPTIMIZER = @adafactor.Adafactor()
31
+ RANDOM_SEED = 0
32
+ TASK_FEATURE_LENGTHS = {'inputs': 512, 'targets': 512}
33
+ TRAIN_STEPS = 1800000
34
+ USE_CACHED_TASKS = False
35
+ USE_HARDWARE_RNG = False
36
+ VOCABULARY = @seqio.SentencePieceVocabulary()
37
+ Z_LOSS = 0.0001
38
+
39
+ # Parameters for adafactor.Adafactor:
40
+ # ==============================================================================
41
+ adafactor.Adafactor.decay_rate = 0.8
42
+ adafactor.Adafactor.logical_factor_rules = \
43
+ @adafactor.standard_logical_factor_rules()
44
+ adafactor.Adafactor.step_offset = 0
45
+
46
+ # Parameters for utils.CheckpointConfig:
47
+ # ==============================================================================
48
+ utils.CheckpointConfig.restore = @utils.RestoreCheckpointConfig()
49
+ utils.CheckpointConfig.save = @utils.SaveCheckpointConfig()
50
+
51
+ # Parameters for utils.create_learning_rate_scheduler:
52
+ # ==============================================================================
53
+ utils.create_learning_rate_scheduler.base_learning_rate = 0.001
54
+ utils.create_learning_rate_scheduler.factors = 'constant'
55
+ utils.create_learning_rate_scheduler.warmup_steps = 1000
56
+
57
+ # Parameters for infer_eval/utils.DatasetConfig:
58
+ # ==============================================================================
59
+ infer_eval/utils.DatasetConfig.batch_size = %BATCH_SIZE
60
+ infer_eval/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
61
+ infer_eval/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
62
+ infer_eval/utils.DatasetConfig.pack = False
63
+ infer_eval/utils.DatasetConfig.seed = 42
64
+ infer_eval/utils.DatasetConfig.shuffle = False
65
+ infer_eval/utils.DatasetConfig.split = 'validation'
66
+ infer_eval/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
67
+ infer_eval/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
68
+
69
+ # Parameters for train/utils.DatasetConfig:
70
+ # ==============================================================================
71
+ train/utils.DatasetConfig.batch_size = %BATCH_SIZE
72
+ train/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
73
+ train/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
74
+ train/utils.DatasetConfig.pack = True
75
+ train/utils.DatasetConfig.seed = None
76
+ train/utils.DatasetConfig.shuffle = True
77
+ train/utils.DatasetConfig.split = 'train'
78
+ train/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
79
+ train/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
80
+
81
+ # Parameters for train_eval/utils.DatasetConfig:
82
+ # ==============================================================================
83
+ train_eval/utils.DatasetConfig.batch_size = %BATCH_SIZE
84
+ train_eval/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
85
+ train_eval/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
86
+ train_eval/utils.DatasetConfig.pack = True
87
+ train_eval/utils.DatasetConfig.seed = 42
88
+ train_eval/utils.DatasetConfig.shuffle = False
89
+ train_eval/utils.DatasetConfig.split = 'validation'
90
+ train_eval/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
91
+ train_eval/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
92
+
93
+ # Parameters for models.EncoderDecoderModel:
94
+ # ==============================================================================
95
+ models.EncoderDecoderModel.input_vocabulary = %VOCABULARY
96
+ models.EncoderDecoderModel.label_smoothing = %LABEL_SMOOTHING
97
+ models.EncoderDecoderModel.loss_normalizing_factor = %LOSS_NORMALIZING_FACTOR
98
+ models.EncoderDecoderModel.module = @network.Transformer()
99
+ models.EncoderDecoderModel.optimizer_def = %OPTIMIZER
100
+ models.EncoderDecoderModel.output_vocabulary = %VOCABULARY
101
+ models.EncoderDecoderModel.z_loss = %Z_LOSS
102
+
103
+ # Parameters for seqio.Evaluator:
104
+ # ==============================================================================
105
+ seqio.Evaluator.logger_cls = \
106
+ [@seqio.PyLoggingLogger, @seqio.TensorBoardLogger, @seqio.JSONLogger]
107
+ seqio.Evaluator.num_examples = %EVALUATOR_NUM_EXAMPLES
108
+ seqio.Evaluator.use_memory_cache = %EVALUATOR_USE_MEMORY_CACHE
109
+
110
+ # Parameters for seqio.JSONLogger:
111
+ # ==============================================================================
112
+ seqio.JSONLogger.write_n_results = %JSON_WRITE_N_RESULTS
113
+
114
+ # Parameters for partitioning.PjitPartitioner:
115
+ # ==============================================================================
116
+ partitioning.PjitPartitioner.logical_axis_rules = \
117
+ @partitioning.standard_logical_axis_rules()
118
+ partitioning.PjitPartitioner.model_parallel_submesh = None
119
+ partitioning.PjitPartitioner.num_partitions = 1
120
+
121
+ # Parameters for utils.RestoreCheckpointConfig:
122
+ # ==============================================================================
123
+ utils.RestoreCheckpointConfig.dtype = 'float32'
124
+ utils.RestoreCheckpointConfig.mode = 'specific'
125
+ utils.RestoreCheckpointConfig.path = %INITIAL_CHECKPOINT_PATH
126
+
127
+ # Parameters for utils.SaveCheckpointConfig:
128
+ # ==============================================================================
129
+ utils.SaveCheckpointConfig.dtype = 'float32'
130
+ utils.SaveCheckpointConfig.keep = None
131
+ utils.SaveCheckpointConfig.period = 10000
132
+ utils.SaveCheckpointConfig.save_dataset = False
133
+
134
+ # Parameters for seqio.SentencePieceVocabulary:
135
+ # ==============================================================================
136
+ seqio.SentencePieceVocabulary.sentencepiece_model_file = \
137
+ 'gs://t5-data/vocabs/mc4.250000.100extra/sentencepiece.model'
138
+
139
+ # Parameters for network.T5Config:
140
+ # ==============================================================================
141
+ network.T5Config.dropout_rate = %DROPOUT_RATE
142
+ network.T5Config.dtype = 'bfloat16'
143
+ network.T5Config.emb_dim = 768
144
+ network.T5Config.head_dim = 64
145
+ network.T5Config.logits_via_embedding = False
146
+ network.T5Config.mlp_activations = ('gelu', 'linear')
147
+ network.T5Config.mlp_dim = 2048
148
+ network.T5Config.num_decoder_layers = 12
149
+ network.T5Config.num_encoder_layers = 12
150
+ network.T5Config.num_heads = 12
151
+ network.T5Config.vocab_size = 250112
152
+
153
+ # Parameters for train_script.train:
154
+ # ==============================================================================
155
+ train_script.train.checkpoint_cfg = @utils.CheckpointConfig()
156
+ train_script.train.eval_period = 1000
157
+ train_script.train.eval_steps = %EVAL_STEPS
158
+ train_script.train.infer_eval_dataset_cfg = @infer_eval/utils.DatasetConfig()
159
+ train_script.train.inference_evaluator_cls = @seqio.Evaluator
160
+ train_script.train.model = %MODEL
161
+ train_script.train.model_dir = %MODEL_DIR
162
+ train_script.train.partitioner = @partitioning.PjitPartitioner()
163
+ train_script.train.random_seed = %RANDOM_SEED
164
+ train_script.train.summarize_config_fn = @gin_utils.summarize_gin_config
165
+ train_script.train.total_steps = %TRAIN_STEPS
166
+ train_script.train.train_dataset_cfg = @train/utils.DatasetConfig()
167
+ train_script.train.train_eval_dataset_cfg = @train_eval/utils.DatasetConfig()
168
+ train_script.train.trainer_cls = @trainer.Trainer
169
+ train_script.train.use_hardware_rng = %USE_HARDWARE_RNG
170
+
171
+ # Parameters for trainer.Trainer:
172
+ # ==============================================================================
173
+ trainer.Trainer.learning_rate_fn = @utils.create_learning_rate_scheduler()
174
+ trainer.Trainer.num_microbatches = None
175
+
176
+ # Parameters for network.Transformer:
177
+ # ==============================================================================
178
+ network.Transformer.config = @network.T5Config()
config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/home/perk/models/demo-deuncaser-base",
3
+ "architectures": [
4
+ "T5ForConditionalGeneration"
5
+ ],
6
+ "d_ff": 2048,
7
+ "d_kv": 64,
8
+ "d_model": 768,
9
+ "decoder_start_token_id": 0,
10
+ "dropout_rate": 0.1,
11
+ "eos_token_id": 1,
12
+ "feed_forward_proj": "gated-gelu",
13
+ "initializer_factor": 1.0,
14
+ "is_encoder_decoder": true,
15
+ "layer_norm_epsilon": 1e-06,
16
+ "model_type": "t5",
17
+ "num_decoder_layers": 12,
18
+ "num_heads": 12,
19
+ "num_layers": 12,
20
+ "output_past": true,
21
+ "pad_token_id": 0,
22
+ "relative_attention_num_buckets": 32,
23
+ "tie_word_embeddings": false,
24
+ "tokenizer_class": "T5Tokenizer",
25
+ "torch_dtype": "float32",
26
+ "transformers_version": "4.17.0",
27
+ "use_cache": true,
28
+ "vocab_size": 250112
29
+ }
flax_model.msgpack ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06e968385c0d1dbade8052c51335df6313da39a0cf18731f44d85eb8a818a586
3
+ size 2329617315
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:d4cb6890a66404d755091d559221ca11ee52e47c67f3674ea6159b068ac3e550
3
+ size 2329729805
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
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.1652376058.t1v-n-569e1ce4-w-0.282691.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:030b518e38f79fbd8d84f3130df6032b09f2e27217e5d6a1499402bbc92a7fe0
3
+ size 162654
training_eval/deuncaser/events.out.tfevents.1652376059.t1v-n-569e1ce4-w-0.282691.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0d3c56cc25321f3912cd2b2424944a50964e95efcf3f0ffd3165e0b467d3a17
3
+ size 130626