from __gin__ import dynamic_registration
import tasks_v4

import __main__ as train_script
from t5.data import mixtures
from t5x import models
from t5x import partitioning
from t5x import utils

include "t5x/examples/t5/mt5/large.gin"
include "t5x/configs/runs/finetune.gin"

MIXTURE_OR_TASK_NAME = "sentencefix"
TASK_FEATURE_LENGTHS = {"inputs": 256, "targets": 256}
TRAIN_STEPS = 1_200_000  # 1000000 pre-trained steps + 20000 fine-tuning steps.
USE_CACHED_TASKS = False
DROPOUT_RATE = 0.0
RANDOM_SEED = 0

# `LOSS_NORMALIZING_FACTOR`: When fine-tuning a model that was pre-trained
# using Mesh Tensorflow (e.g. the public T5 / mT5 / ByT5 models), this should be
# set to `pretraining batch_size` * `target_token_length`. For T5 and T5.1.1:
# `2048 * 114`. For mT5: `1024 * 229`. For ByT5: `1024 * 189`.
#LOSS_NORMALIZING_FACTOR = 234496
INITIAL_CHECKPOINT_PATH = "gs://t5-data/pretrained_models/t5x/mt5_large/checkpoint_1000000"

train_script.train:
  eval_period = 100
  partitioner = @partitioning.PjitPartitioner()

utils.RestoreCheckpointConfig:
  path = %INITIAL_CHECKPOINT_PATH
  mode = 'specific'

#train_script.train:
#  train_dataset_cfg = @train/utils.DatasetConfig()
#  train_eval_dataset_cfg = @train_eval/utils.DatasetConfig()
#  infer_eval_dataset_cfg = @infer_eval/utils.DatasetConfig()

models.EncoderDecoderModel.predict_batch_with_aux.num_decodes = 4

infer_eval/utils.DatasetConfig:
  mixture_or_task_name = %MIXTURE_OR_TASK_NAME
  task_feature_lengths = %TASK_FEATURE_LENGTHS
  split = 'validation'
  batch_size = 64
  shuffle = False
  seed = 42
  use_cached = %USE_CACHED_TASKS
  pack = False
  module = %MIXTURE_OR_TASK_MODULE

partitioning.PjitPartitioner:
  num_partitions = 4
  model_parallel_submesh = None
  logical_axis_rules = @partitioning.standard_logical_axis_rules()

#partitioning.PjitPartitioner.num_partitions = 4