pere commited on
Commit
3dd9379
1 Parent(s): c97a733

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 +1,151 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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/t5x/mt5_small/checkpoint_1000000'
20
+ LABEL_SMOOTHING = 0.0
21
+ LOSS_NORMALIZING_FACTOR = None
22
+ MIXTURE_OR_TASK_MODULE = None
23
+ MIXTURE_OR_TASK_NAME = 'ncc_english_span_corruption_stream'
24
+ MODEL = @models.EncoderDecoderModel()
25
+ MODEL_DIR = 'gs://nb-t5x-us-central2/norwegian_NCC_plus_English_t5x_small'
26
+ OPTIMIZER = @adafactor.Adafactor()
27
+ RANDOM_SEED = None
28
+ SHUFFLE_TRAIN_EXAMPLES = True
29
+ TASK_FEATURE_LENGTHS = {'inputs': 512, 'targets': 512}
30
+ TRAIN_STEPS = 1500000
31
+ USE_CACHED_TASKS = True
32
+ USE_HARDWARE_RNG = False
33
+ VOCABULARY = @seqio.SentencePieceVocabulary()
34
+ Z_LOSS = 0.0001
35
+
36
+ # Parameters for adafactor.Adafactor:
37
+ # ==============================================================================
38
+ adafactor.Adafactor.decay_rate = 0.8
39
+ adafactor.Adafactor.logical_factor_rules = \
40
+ @adafactor.standard_logical_factor_rules()
41
+ adafactor.Adafactor.step_offset = 0
42
+
43
+ # Parameters for utils.CheckpointConfig:
44
+ # ==============================================================================
45
+ utils.CheckpointConfig.restore = @utils.RestoreCheckpointConfig()
46
+ utils.CheckpointConfig.save = @utils.SaveCheckpointConfig()
47
+
48
+ # Parameters for utils.create_learning_rate_scheduler:
49
+ # ==============================================================================
50
+ utils.create_learning_rate_scheduler.base_learning_rate = 0.5
51
+ utils.create_learning_rate_scheduler.factors = 'constant * rsqrt_decay'
52
+ utils.create_learning_rate_scheduler.warmup_steps = 10000
53
+
54
+ # Parameters for train/utils.DatasetConfig:
55
+ # ==============================================================================
56
+ train/utils.DatasetConfig.batch_size = %BATCH_SIZE
57
+ train/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
58
+ train/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
59
+ train/utils.DatasetConfig.pack = True
60
+ train/utils.DatasetConfig.seed = None
61
+ train/utils.DatasetConfig.shuffle = %SHUFFLE_TRAIN_EXAMPLES
62
+ train/utils.DatasetConfig.split = 'train'
63
+ train/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
64
+ train/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
65
+
66
+ # Parameters for train_eval/utils.DatasetConfig:
67
+ # ==============================================================================
68
+ train_eval/utils.DatasetConfig.batch_size = %BATCH_SIZE
69
+ train_eval/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
70
+ train_eval/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
71
+ train_eval/utils.DatasetConfig.pack = True
72
+ train_eval/utils.DatasetConfig.seed = 42
73
+ train_eval/utils.DatasetConfig.shuffle = False
74
+ train_eval/utils.DatasetConfig.split = 'validation'
75
+ train_eval/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
76
+ train_eval/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
77
+
78
+ # Parameters for models.EncoderDecoderModel:
79
+ # ==============================================================================
80
+ models.EncoderDecoderModel.input_vocabulary = %VOCABULARY
81
+ models.EncoderDecoderModel.label_smoothing = %LABEL_SMOOTHING
82
+ models.EncoderDecoderModel.loss_normalizing_factor = %LOSS_NORMALIZING_FACTOR
83
+ models.EncoderDecoderModel.module = @network.Transformer()
84
+ models.EncoderDecoderModel.optimizer_def = %OPTIMIZER
85
+ models.EncoderDecoderModel.output_vocabulary = %VOCABULARY
86
+ models.EncoderDecoderModel.z_loss = %Z_LOSS
87
+
88
+ # Parameters for partitioning.PjitPartitioner:
89
+ # ==============================================================================
90
+ partitioning.PjitPartitioner.logical_axis_rules = \
91
+ @partitioning.standard_logical_axis_rules()
92
+ partitioning.PjitPartitioner.model_parallel_submesh = None
93
+ partitioning.PjitPartitioner.num_partitions = 2
94
+
95
+ # Parameters for utils.RestoreCheckpointConfig:
96
+ # ==============================================================================
97
+ utils.RestoreCheckpointConfig.dtype = 'float32'
98
+ utils.RestoreCheckpointConfig.mode = 'specific'
99
+ utils.RestoreCheckpointConfig.path = %INITIAL_CHECKPOINT_PATH
100
+
101
+ # Parameters for utils.SaveCheckpointConfig:
102
+ # ==============================================================================
103
+ utils.SaveCheckpointConfig.dtype = 'float32'
104
+ utils.SaveCheckpointConfig.keep = None
105
+ utils.SaveCheckpointConfig.period = 1000
106
+ utils.SaveCheckpointConfig.save_dataset = False
107
+
108
+ # Parameters for seqio.SentencePieceVocabulary:
109
+ # ==============================================================================
110
+ seqio.SentencePieceVocabulary.sentencepiece_model_file = \
111
+ 'gs://t5-data/vocabs/mc4.250000.100extra/sentencepiece.model'
112
+
113
+ # Parameters for network.T5Config:
114
+ # ==============================================================================
115
+ network.T5Config.dropout_rate = %DROPOUT_RATE
116
+ network.T5Config.dtype = 'bfloat16'
117
+ network.T5Config.emb_dim = 512
118
+ network.T5Config.head_dim = 64
119
+ network.T5Config.logits_via_embedding = False
120
+ network.T5Config.mlp_activations = ('gelu', 'linear')
121
+ network.T5Config.mlp_dim = 1024
122
+ network.T5Config.num_decoder_layers = 8
123
+ network.T5Config.num_encoder_layers = 8
124
+ network.T5Config.num_heads = 6
125
+ network.T5Config.vocab_size = 250112
126
+
127
+ # Parameters for train_script.train:
128
+ # ==============================================================================
129
+ train_script.train.checkpoint_cfg = @utils.CheckpointConfig()
130
+ train_script.train.eval_period = 1000
131
+ train_script.train.eval_steps = 20
132
+ train_script.train.infer_eval_dataset_cfg = None
133
+ train_script.train.model = %MODEL
134
+ train_script.train.model_dir = %MODEL_DIR
135
+ train_script.train.partitioner = @partitioning.PjitPartitioner()
136
+ train_script.train.random_seed = %RANDOM_SEED
137
+ train_script.train.summarize_config_fn = @gin_utils.summarize_gin_config
138
+ train_script.train.total_steps = %TRAIN_STEPS
139
+ train_script.train.train_dataset_cfg = @train/utils.DatasetConfig()
140
+ train_script.train.train_eval_dataset_cfg = @train_eval/utils.DatasetConfig()
141
+ train_script.train.trainer_cls = @trainer.Trainer
142
+ train_script.train.use_hardware_rng = %USE_HARDWARE_RNG
143
+
144
+ # Parameters for trainer.Trainer:
145
+ # ==============================================================================
146
+ trainer.Trainer.learning_rate_fn = @utils.create_learning_rate_scheduler()
147
+ trainer.Trainer.num_microbatches = None
148
+
149
+ # Parameters for network.Transformer:
150
+ # ==============================================================================
151
+ network.Transformer.config = @network.T5Config()
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "T5ForConditionalGeneration"
4
+ ],
5
+ "d_ff": 1024,
6
+ "d_kv": 64,
7
+ "d_model": 512,
8
+ "decoder_start_token_id": 0,
9
+ "dropout_rate": 0.1,
10
+ "eos_token_id": 1,
11
+ "feed_forward_proj": "gated-gelu",
12
+ "initializer_factor": 1.0,
13
+ "is_encoder_decoder": true,
14
+ "layer_norm_epsilon": 1e-06,
15
+ "model_type": "t5",
16
+ "num_decoder_layers": 8,
17
+ "num_heads": 6,
18
+ "num_layers": 8,
19
+ "pad_token_id": 0,
20
+ "relative_attention_max_distance": 128,
21
+ "relative_attention_num_buckets": 32,
22
+ "tie_word_embeddings": false,
23
+ "tokenizer_class": "T5Tokenizer",
24
+ "torch_dtype": "float32",
25
+ "transformers_version": "4.19.2",
26
+ "use_cache": true,
27
+ "vocab_size": 250112
28
+ }
flax_model.msgpack ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:530626e8dfa10d0f212f954a78b44b9096313537e68dbe56e5a84a8735c514c4
3
+ size 1200715307
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 sha256:6313905e2cde2b2a029d273148c8c306e60da5625e45864a331a9efa889db865
3
+ size 1200768197
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 sha256:13edf9346263c231557abbd4b42fd847571a957f938e4f48b2d23ec2f1c58acc
3
+ size 9760
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93c3578052e1605d8332eb961bc08d72e246071974e4cc54aa6991826b802aa5
3
+ size 16330369
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", "name_or_path": "/home/perk/models/t5_small_NCC", "sp_model_kwargs": {}, "tokenizer_class": "T5Tokenizer"}
train/events.out.tfevents.1649404635.t1v-n-b482f629-w-0.619382.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:250f2c628947c43a81d6c3802b2bb5cb17c2ba55c81f05c0c4a82a70c398b16d
3
+ size 121897
train/events.out.tfevents.1649436246.t1v-n-b482f629-w-0.817332.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4799c96972cc2545e4bc66575db9ccdefd3227ffcff940a5400373e06b427b15
3
+ size 209677
train/events.out.tfevents.1649496459.t1v-n-b482f629-w-0.1167309.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7d228ecb1ae90b28f463b139c5b2ee81467c356f6f05e81741458f82ffe9410
3
+ size 121897
train/events.out.tfevents.1649527281.t1v-n-b482f629-w-0.1364734.0.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b240fff366049dad215e9a78b3be800a3d3654ac5e9ffd522ba2a62b22ee0e41
3
+ size 307783
training_eval/ncc_english_span_corruption_stream/events.out.tfevents.1649404635.t1v-n-b482f629-w-0.619382.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f4e0d4d7f376cb266f6739756ab545fc4d4c1702cd14503cdf44a0b61114d91
3
+ size 101916
training_eval/ncc_english_span_corruption_stream/events.out.tfevents.1649436246.t1v-n-b482f629-w-0.817332.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f78548ed75a88df9dec873d078fac2a43bc1e380ea9d1449d48dc0acf3af5ed
3
+ size 180216
training_eval/ncc_english_span_corruption_stream/events.out.tfevents.1649496459.t1v-n-b482f629-w-0.1167309.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e63d51c3af44ccaa16c15400e69ad89bb8772192a04543587e23bbf6217f6d04
3
+ size 101916
training_eval/ncc_english_span_corruption_stream/events.out.tfevents.1649527281.t1v-n-b482f629-w-0.1364734.1.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e81dfcfb8841a98f1c1e15c24f4cd9a2d59138e00a1ab93e6d34f5bf21e42dc7
3
+ size 268956