first epoch

first_epoch/config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "outputs",
+  "architectures": [
+    "MT5ForConditionalGeneration"
+  ],
+  "d_ff": 1024,
+  "d_kv": 64,
+  "d_model": 512,
+  "decoder_start_token_id": 0,
+  "dense_act_fn": "gelu_new",
+  "dropout_rate": 0.1,
+  "eos_token_id": 1,
+  "feed_forward_proj": "gated-gelu",
+  "initializer_factor": 1.0,
+  "is_encoder_decoder": true,
+  "is_gated_act": true,
+  "layer_norm_epsilon": 1e-06,
+  "model_type": "mt5",
+  "num_decoder_layers": 8,
+  "num_heads": 6,
+  "num_layers": 8,
+  "pad_token_id": 0,
+  "relative_attention_max_distance": 128,
+  "relative_attention_num_buckets": 32,
+  "tie_word_embeddings": false,
+  "tokenizer_class": "T5Tokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.27.4",
+  "use_cache": true,
+  "vocab_size": 250112
+}

first_epoch/eval_results.txt

@@ -0,0 +1 @@
+eval_loss = 2.8873880844224584

first_epoch/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "decoder_start_token_id": 0,
+  "eos_token_id": 1,
+  "pad_token_id": 0,
+  "transformers_version": "4.27.4"
+}

first_epoch/model_args.json

@@ -0,0 +1 @@
+{"adafactor_beta1": null, "adafactor_clip_threshold": 1.0, "adafactor_decay_rate": -0.8, "adafactor_eps": [1e-30, 0.001], "adafactor_relative_step": false, "adafactor_scale_parameter": false, "adafactor_warmup_init": false, "adam_betas": [0.9, 0.999], "adam_epsilon": 1e-08, "best_model_dir": "outputs/best_model", "cache_dir": "cache_dir/", "config": {}, "cosine_schedule_num_cycles": 0.5, "custom_layer_parameters": [], "custom_parameter_groups": [], "dataloader_num_workers": 0, "do_lower_case": false, "dynamic_quantize": false, "early_stopping_consider_epochs": false, "early_stopping_delta": 0, "early_stopping_metric": "eval_loss", "early_stopping_metric_minimize": true, "early_stopping_patience": 3, "encoding": null, "eval_batch_size": 20, "evaluate_during_training": true, "evaluate_during_training_silent": true, "evaluate_during_training_steps": 200, "evaluate_during_training_verbose": false, "evaluate_each_epoch": true, "fp16": false, "gradient_accumulation_steps": 1, "learning_rate": 0.001, "local_rank": -1, "logging_steps": 50, "loss_type": null, "loss_args": {}, "manual_seed": null, "max_grad_norm": 1.0, "max_seq_length": 96, "model_name": "outputs", "model_type": "mt5", "multiprocessing_chunksize": -1, "n_gpu": 1, "no_cache": true, "no_save": false, "not_saved_args": [], "num_train_epochs": 4, "optimizer": "Adafactor", "output_dir": "outputs/", "overwrite_output_dir": false, "polynomial_decay_schedule_lr_end": 1e-07, "polynomial_decay_schedule_power": 1.0, "process_count": 2, "quantized_model": false, "reprocess_input_data": true, "save_best_model": true, "save_eval_checkpoints": false, "save_model_every_epoch": true, "save_optimizer_and_scheduler": true, "save_steps": -1, "scheduler": "constant_schedule_with_warmup", "silent": false, "skip_special_tokens": true, "tensorboard_dir": null, "thread_count": null, "tokenizer_name": null, "tokenizer_type": null, "train_batch_size": 20, "train_custom_parameters_only": false, "use_cached_eval_features": false, "use_early_stopping": false, "use_hf_datasets": false, "use_multiprocessing": false, "use_multiprocessing_for_evaluation": true, "wandb_kwargs": {}, "wandb_project": "MT5 English-Chinese Translation", "warmup_ratio": 0.06, "warmup_steps": 5551, "weight_decay": 0.0, "model_class": "T5Model", "dataset_class": null, "do_sample": false, "early_stopping": true, "evaluate_generated_text": false, "length_penalty": 2.0, "max_length": 20, "max_steps": -1, "num_beams": 1, "num_return_sequences": 1, "preprocess_inputs": false, "repetition_penalty": 1.0, "special_tokens_list": [], "top_k": null, "top_p": null, "use_multiprocessed_decoding": true}

first_epoch/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc41bbe83da72953f4a81cb7c890d2e72173ffbf78b53c99bb921f8de97089e8
+size 2879429

first_epoch/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b70817e39971bcc7b3fa896c7ccb218f90d4ab826f6c22d91e6acccd6da7501d
+size 1200772485

first_epoch/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:439a6cd8a3b103eaafbd3c0ef26fb99265c72def63ccb9ca9b250e82a1d9d11c
+size 627

first_epoch/special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

first_epoch/spiece.model

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

first_epoch/tokenizer_config.json

@@ -0,0 +1,12 @@
+{
+  "additional_special_tokens": null,
+  "eos_token": "</s>",
+  "extra_ids": 0,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "sp_model_kwargs": {},
+  "special_tokens_map_file": "/home/ec2-user/.cache/huggingface/hub/models--google--mt5-small/snapshots/38f23af8ec210eb6c376d40e9c56bd25a80f195d/special_tokens_map.json",
+  "tokenizer_class": "T5Tokenizer",
+  "truncate": true,
+  "unk_token": "<unk>"
+}

first_epoch/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98a8c70371bd2f02f5ec3c8e670ec3c404c1635958b16e861bc0a61c11beefc5
+size 3195

test.py

@@ -0,0 +1,33 @@
+from datasets.load import load_dataset
+import logging
+import sacrebleu
+import pandas as pd
+from simpletransformers.t5 import T5Model, T5Args
+
+raw_datasets = load_dataset('iwslt2017', 'iwslt2017-zh-en')
+
+logging.basicConfig(level=logging.INFO)
+transformers_logger = logging.getLogger("transformers")
+transformers_logger.setLevel(logging.WARNING)
+
+
+model_args = T5Args()
+model_args.max_length = 512
+model_args.length_penalty = 1
+model_args.num_beams = 10
+
+model = T5Model("mt5", "outputs", args=model_args)
+
+en_zh_test = pd.DataFrame(raw_datasets['test']['translation'])
+zh_truth = en_zh_test['zh'].tolist()
+en_input = en_zh_test['en'].tolist()
+
+zh_preds = model.predict(en_input)
+en_zh_bleu = sacrebleu.corpus_bleu(zh_preds, zh_truth)
+print("----------------------------------------------")
+print("English to Chinese: ", en_zh_bleu.score)
+
+en_preds = model.predict(zh_truth)
+zh_en_bleu = sacrebleu.corpus_bleu(en_preds, en_input)
+print("----------------------------------------------")
+print("Chinese to English: ", zh_en_bleu.score)

train.py

@@ -0,0 +1,49 @@
+from datasets.load import load_dataset
+import pandas as pd
+import logging
+from simpletransformers.t5 import T5Args, T5Model
+
+logging.basicConfig(level=logging.INFO)
+transformers_logger = logging.getLogger("transformers")
+transformers_logger.setLevel(logging.WARNING)
+
+raw_datasets = load_dataset('iwslt2017', 'iwslt2017-zh-en')
+
+train_df = pd.DataFrame(raw_datasets['train']['translation'])
+train_df.columns = ['input_text', 'target_text']
+reverse_df = train_df.copy()
+reverse_df.columns = ['target_text', 'input_text']
+train_df['prefix'] = 'translate english to chinese'
+reverse_df['prefix'] = 'translate chinese to english'
+train_df = pd.concat([train_df, reverse_df])
+
+eval_df = pd.DataFrame(raw_datasets['validation']['translation'])
+eval_df.columns = ['input_text', 'target_text']
+reverse_df = eval_df.copy()
+reverse_df.columns = ['target_text', 'input_text']
+eval_df['prefix'] = 'translate english to chinese'
+reverse_df['prefix'] = 'translate chinese to english'
+eval_df = pd.concat([eval_df, reverse_df])
+
+model_args = T5Args()
+model_args.max_seq_length = 96
+model_args.train_batch_size = 20
+model_args.eval_batch_size = 20
+model_args.num_train_epochs = 4
+model_args.evaluate_during_training = True
+model_args.evaluate_during_training_steps = 5000
+model_args.use_multiprocessing = False
+model_args.fp16 = False
+model_args.save_steps = -1
+model_args.save_model_every_epoch = True
+model_args.save_eval_checkpoints = False
+model_args.no_cache = True
+model_args.reprocess_input_data = True
+model_args.overwrite_output_dir = False
+model_args.preprocess_inputs = False
+model_args.num_return_sequences = 1
+model_args.wandb_project = "MT5 English-Chinese Translation"
+
+model = T5Model("mt5", "outputs", args=model_args)
+
+model.train_model(train_df, eval_data=eval_df, output_dir='mt5_more_epochs')