Commit from model create scripts

.gitattributes

@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

config.gin

@@ -0,0 +1,147 @@
+from __gin__ import dynamic_registration
+import __main__ as train_script
+import seqio
+import t5.data.mixtures
+from t5x import adafactor
+from t5x.examples.t5 import network
+from t5x import gin_utils
+from t5x import models
+from t5x import partitioning
+from t5x import trainer
+from t5x import utils
+import tasks
+
+# Macros:
+# ==============================================================================
+BATCH_SIZE = 128
+DROPOUT_RATE = 0.0
+INITIAL_CHECKPOINT_PATH = \
+    'gs://t5-data/pretrained_models/byt5/base/model.ckpt-1000000'
+LABEL_SMOOTHING = 0.0
+LOSS_NORMALIZING_FACTOR = 193536
+MIXTURE_OR_TASK_MODULE = None
+MIXTURE_OR_TASK_NAME = 'byt5_ncc_english_span_corruption_stream'
+MODEL = @models.EncoderDecoderModel()
+MODEL_DIR = \
+    'gs://t5x-training/pretrained_models/norwegian_NCC_plus_English_byt5x_base'
+OPTIMIZER = @adafactor.Adafactor()
+RANDOM_SEED = None
+SHUFFLE_TRAIN_EXAMPLES = True
+TASK_FEATURE_LENGTHS = {'inputs': 512, 'targets': 512}
+TRAIN_STEPS = 1500000
+USE_CACHED_TASKS = True
+USE_HARDWARE_RNG = False
+VOCABULARY = @seqio.ByteVocabulary()
+Z_LOSS = 0.0001
+
+# Parameters for adafactor.Adafactor:
+# ==============================================================================
+adafactor.Adafactor.decay_rate = 0.8
+adafactor.Adafactor.logical_factor_rules = \
+    @adafactor.standard_logical_factor_rules()
+adafactor.Adafactor.step_offset = 0
+
+# Parameters for utils.CheckpointConfig:
+# ==============================================================================
+utils.CheckpointConfig.restore = @utils.RestoreCheckpointConfig()
+utils.CheckpointConfig.save = @utils.SaveCheckpointConfig()
+
+# Parameters for utils.create_learning_rate_scheduler:
+# ==============================================================================
+utils.create_learning_rate_scheduler.base_learning_rate = 0.5
+utils.create_learning_rate_scheduler.factors = 'constant * rsqrt_decay'
+utils.create_learning_rate_scheduler.warmup_steps = 10000
+
+# Parameters for train/utils.DatasetConfig:
+# ==============================================================================
+train/utils.DatasetConfig.batch_size = %BATCH_SIZE
+train/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
+train/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
+train/utils.DatasetConfig.pack = True
+train/utils.DatasetConfig.seed = None
+train/utils.DatasetConfig.shuffle = %SHUFFLE_TRAIN_EXAMPLES
+train/utils.DatasetConfig.split = 'train'
+train/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
+train/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
+
+# Parameters for train_eval/utils.DatasetConfig:
+# ==============================================================================
+train_eval/utils.DatasetConfig.batch_size = %BATCH_SIZE
+train_eval/utils.DatasetConfig.mixture_or_task_name = %MIXTURE_OR_TASK_NAME
+train_eval/utils.DatasetConfig.module = %MIXTURE_OR_TASK_MODULE
+train_eval/utils.DatasetConfig.pack = True
+train_eval/utils.DatasetConfig.seed = 42
+train_eval/utils.DatasetConfig.shuffle = False
+train_eval/utils.DatasetConfig.split = 'validation'
+train_eval/utils.DatasetConfig.task_feature_lengths = %TASK_FEATURE_LENGTHS
+train_eval/utils.DatasetConfig.use_cached = %USE_CACHED_TASKS
+
+# Parameters for models.EncoderDecoderModel:
+# ==============================================================================
+models.EncoderDecoderModel.input_vocabulary = %VOCABULARY
+models.EncoderDecoderModel.label_smoothing = %LABEL_SMOOTHING
+models.EncoderDecoderModel.loss_normalizing_factor = %LOSS_NORMALIZING_FACTOR
+models.EncoderDecoderModel.module = @network.Transformer()
+models.EncoderDecoderModel.optimizer_def = %OPTIMIZER
+models.EncoderDecoderModel.output_vocabulary = %VOCABULARY
+models.EncoderDecoderModel.z_loss = %Z_LOSS
+
+# Parameters for partitioning.PjitPartitioner:
+# ==============================================================================
+partitioning.PjitPartitioner.logical_axis_rules = \
+    @partitioning.standard_logical_axis_rules()
+partitioning.PjitPartitioner.model_parallel_submesh = None
+partitioning.PjitPartitioner.num_partitions = 1
+
+# Parameters for utils.RestoreCheckpointConfig:
+# ==============================================================================
+utils.RestoreCheckpointConfig.dtype = 'float32'
+utils.RestoreCheckpointConfig.mode = 'specific'
+utils.RestoreCheckpointConfig.path = %INITIAL_CHECKPOINT_PATH
+
+# Parameters for utils.SaveCheckpointConfig:
+# ==============================================================================
+utils.SaveCheckpointConfig.dtype = 'float32'
+utils.SaveCheckpointConfig.keep = None
+utils.SaveCheckpointConfig.period = 1000
+utils.SaveCheckpointConfig.save_dataset = False
+
+# Parameters for network.T5Config:
+# ==============================================================================
+network.T5Config.dropout_rate = %DROPOUT_RATE
+network.T5Config.dtype = 'bfloat16'
+network.T5Config.emb_dim = 1536
+network.T5Config.head_dim = 64
+network.T5Config.logits_via_embedding = False
+network.T5Config.mlp_activations = ('gelu', 'linear')
+network.T5Config.mlp_dim = 3968
+network.T5Config.num_decoder_layers = 6
+network.T5Config.num_encoder_layers = 18
+network.T5Config.num_heads = 12
+network.T5Config.vocab_size = 384
+
+# Parameters for train_script.train:
+# ==============================================================================
+train_script.train.checkpoint_cfg = @utils.CheckpointConfig()
+train_script.train.eval_period = 1000
+train_script.train.eval_steps = 20
+train_script.train.infer_eval_dataset_cfg = None
+train_script.train.model = %MODEL
+train_script.train.model_dir = %MODEL_DIR
+train_script.train.partitioner = @partitioning.PjitPartitioner()
+train_script.train.random_seed = %RANDOM_SEED
+train_script.train.summarize_config_fn = @gin_utils.summarize_gin_config
+train_script.train.total_steps = %TRAIN_STEPS
+train_script.train.train_dataset_cfg = @train/utils.DatasetConfig()
+train_script.train.train_eval_dataset_cfg = @train_eval/utils.DatasetConfig()
+train_script.train.trainer_cls = @trainer.Trainer
+train_script.train.use_hardware_rng = %USE_HARDWARE_RNG
+
+# Parameters for trainer.Trainer:
+# ==============================================================================
+trainer.Trainer.learning_rate_fn = @utils.create_learning_rate_scheduler()
+trainer.Trainer.num_microbatches = None
+
+# Parameters for network.Transformer:
+# ==============================================================================
+network.Transformer.config = @network.T5Config()

config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "/home/perk/models/byt5_base_NCC",
+  "architectures": [
+    "T5ForConditionalGeneration"
+  ],
+  "d_ff": 3968,
+  "d_kv": 64,
+  "d_model": 1536,
+  "decoder_start_token_id": 0,
+  "dropout_rate": 0.1,
+  "eos_token_id": 1,
+  "feed_forward_proj": "gated-gelu",
+  "gradient_checkpointing": false,
+  "initializer_factor": 1.0,
+  "is_encoder_decoder": true,
+  "layer_norm_epsilon": 1e-06,
+  "model_type": "t5",
+  "num_decoder_layers": 6,
+  "num_heads": 12,
+  "num_layers": 18,
+  "output_past": true,
+  "pad_token_id": 0,
+  "relative_attention_max_distance": 128,
+  "relative_attention_num_buckets": 32,
+  "tie_word_embeddings": false,
+  "tokenizer_class": "ByT5Tokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.19.2",
+  "use_cache": true,
+  "vocab_size": 384
+}

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:055bba8e7d7e936a3e563a9ddabe78e4c236e13dc127c1b9c55deeb57c00cebe
+size 2326623947

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90728b46881d00abdea86a164bf4c959a88b8f90462a0810bed6e20567af9fd3
+size 2326694409

special_tokens_map.json

@@ -0,0 +1 @@
+{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

spiece.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef78f86560d809067d12bac6c09f19a462cb3af3f54d2b8acbba26e1433125d6
+size 4309802

tf_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:968e4dfd9da58aaad02eca7656a472c46f981be3d39f1fb29cc93e7eae8a4b92
+size 2327127104

tokenizer_config.json

@@ -0,0 +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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e1612e32da33d510bbcbb0cc0e918d3c00d39e33b5906d8c2a67709b2934dbc
+size 6146

train/events.out.tfevents.1651948631.t1v-n-690df9bb-w-0.2371380.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f58e28c4d7cf0b28ed2ff2a76b7af5b6a5885ad43fd8af6ac4e32c67882a1531
+size 84976

train/events.out.tfevents.1652004005.t1v-n-690df9bb-w-0.2538803.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2da4f6a4b431fb45ddcc99d5387e1c050851a153a9bce66c207d774bbec254db
+size 494136

train/events.out.tfevents.1652278024.t1v-n-690df9bb-w-0.23378.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7bc69901de0aadfc39c509ee176bfc8f8bd9ada5857e1b33ca67b5e2859356dd
+size 58728

train/events.out.tfevents.1652333049.t1v-n-eaf6c591-w-0.10368.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30d1798b6d7d4fb0b865c719567a5fab13c4f714d7c8765d2727629a625ed295
+size 159088

training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1651948512.t1v-n-690df9bb-w-0.2369704.1.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14363ddea0c625ec6f45939c124e6fdc0050a43564f5dd2423bd83aa5878e5cb
+size 40

training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1651948632.t1v-n-690df9bb-w-0.2371380.1.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02f4ca66df4cb44578b0af1ffcd1168ac8b314753e57ab6980ed9c34e5b6ad2d
+size 66681

training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652004005.t1v-n-690df9bb-w-0.2538803.1.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24325cb7f945e38d476f50c340ba03f21458fdbcc3ed607899902df07c0d02da
+size 412506

training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652278024.t1v-n-690df9bb-w-0.23378.1.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57fe574e64c5e25055cbee94289c39c743ddca9c6f6a65dd47212c478dcaa00f
+size 44496

training_eval/byt5_ncc_english_span_corruption_stream/events.out.tfevents.1652333050.t1v-n-eaf6c591-w-0.10368.1.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccd5ee8d99d32ae03f37b55c3ce500a0f79152ae2847b283e26c5582f9a55345
+size 129321