main scripts ready for training. No tokenizer yet.

__pycache__/t5_tokenizer_model.cpython-38.pyc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5fe110f60beb470a8651b6f08ad0fa70ed76f09433a9952520db00965485dbc
+size 3425

config.json

@@ -0,0 +1,28 @@
+{
+  "_name_or_path": "/home/patrick/hugging_face/t5/t5-v1_1-base",
+  "architectures": [
+    "T5ForConditionalGeneration"
+  ],
+  "d_ff": 2048,
+  "d_kv": 64,
+  "d_model": 768,
+  "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": 12,
+  "num_heads": 12,
+  "num_layers": 12,
+  "output_past": true,
+  "pad_token_id": 0,
+  "relative_attention_num_buckets": 32,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.10.0.dev0",
+  "use_cache": true,
+  "vocab_size": 50000
+}

run.sh

@@ -0,0 +1,25 @@
+./run_t5_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="t5" \
+    --config_name="./" \
+    --tokenizer_name="./" \
+    --train_file /mnt/disks/flaxdisk/corpus/norwegian_colossal_corpus_train.json \
+    --validation_file /mnt/disks/flaxdisk/corpus/norwegian_colossal_corpus_validation.json \
+    --max_seq_length="128" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="128" \
+    --per_device_eval_batch_size="128" \
+    --learning_rate="8e-3" \
+    --warmup_steps="2000" \
+    --overwrite_output_dir \
+    --cache_dir /mnt/disks/flaxdisk/cache/ \
+    --num_train_epochs="3" \
+    --adam_beta1="0.9" \
+    --adam_beta2="0.98" \
+    --logging_steps="100" \
+    --save_steps="2500" \
+    --eval_steps="2500" \
+    --preprocessing_num_workers 96 \
+    --adafactor \
+    --push_to_hub
+

t5_tokenizer_model.py

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b4fa09755fb0bc80e4fdd49742cd0d5aed816ceb6cd91fe796420b32752b521
+size 3883

train_tokenizer.py

@@ -0,0 +1,34 @@
+from datasets import load_dataset, concatenate_datasets
+from tokenizers import trainers, Tokenizer, normalizers
+from t5_tokenizer_model import SentencePieceUnigramTokenizer
+
+
+vocab_size = 50_000
+input_sentence_size = None
+model_dir = "./"  # ${MODEL_DIR}
+
+# Initialize a dataset
+dataset = load_dataset("json", data_files=["/mnt/disks/flaxdisk/corpus/norwegian_colossal_corpus_validation.json","/mnt/disks/flaxdisk/corpus/special_chars.json"], split='train')
+
+tokenizer = SentencePieceUnigramTokenizer(unk_token="<unk>", eos_token="</s>", pad_token="<pad>")
+
+
+# Build an iterator over this dataset
+def batch_iterator(input_sentence_size=None):
+    if input_sentence_size is None:
+        input_sentence_size = len(dataset)
+    batch_length = 100
+    for i in range(0, input_sentence_size, batch_length):
+        yield dataset[i: i + batch_length]["text"]
+
+
+# Train tokenizer
+tokenizer.train_from_iterator(
+    iterator=batch_iterator(input_sentence_size=input_sentence_size),
+    vocab_size=vocab_size,
+    show_progress=True,
+)
+
+# Save files to disk
+tokenizer.save(f"{model_dir}/tokenizer.json")
+