初始化项目，由ModelHub XC社区提供模型

Model: flax-community/gpt2-medium-indonesian
Source: Original Platform

.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+wandb/

README.md

@@ -0,0 +1,141 @@
+---
+language: id
+widget:
+- text: "Sewindu sudah kita tak berjumpa, rinduku padamu sudah tak terkira."
+---
+
+# GPT2-medium-indonesian
+
+This is a pretrained model on Indonesian language using a causal language modeling (CLM) objective, which was first 
+introduced in [this paper](https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf)
+and first released at [this page](https://openai.com/blog/better-language-models/). 
+
+This model was trained using HuggingFace's Flax framework and is part of the [JAX/Flax Community Week](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104)
+organized by [HuggingFace](https://huggingface.co). All training was done on a TPUv3-8 VM sponsored by the Google Cloud team.
+
+The demo can be found [here](https://huggingface.co/spaces/flax-community/gpt2-indonesian).
+
+## How to use
+You can use this model directly with a pipeline for text generation. Since the generation relies on some randomness, we set a seed for reproducibility:
+```python
+>>> from transformers import pipeline, set_seed
+>>> generator = pipeline('text-generation', model='flax-community/gpt2-medium-indonesian')
+>>> set_seed(42)
+>>> generator("Sewindu sudah kita tak berjumpa,", max_length=30, num_return_sequences=5)
+
+[{'generated_text': 'Sewindu sudah kita tak berjumpa, dua dekade lalu, saya hanya bertemu sekali. Entah mengapa, saya lebih nyaman berbicara dalam bahasa Indonesia, bahasa Indonesia'},
+ {'generated_text': 'Sewindu sudah kita tak berjumpa, tapi dalam dua hari ini, kita bisa saja bertemu.”\
+“Kau tau, bagaimana dulu kita bertemu?” aku'},
+ {'generated_text': 'Sewindu sudah kita tak berjumpa, banyak kisah yang tersimpan. Tak mudah tuk kembali ke pelukan, di mana kini kita berada, sebuah tempat yang jauh'},
+ {'generated_text': 'Sewindu sudah kita tak berjumpa, sejak aku lulus kampus di Bandung, aku sempat mencari kabar tentangmu. Ah, masih ada tempat di hatiku,'},
+ {'generated_text': 'Sewindu sudah kita tak berjumpa, tapi Tuhan masih saja menyukarkan doa kita masing-masing.\
+Tuhan akan memberi lebih dari apa yang kita'}]
+```
+
+Here is how to use this model to get the features of a given text in PyTorch:
+```python
+from transformers import GPT2Tokenizer, GPT2Model
+tokenizer = GPT2Tokenizer.from_pretrained('flax-community/gpt2-medium-indonesian')
+model = GPT2Model.from_pretrained('flax-community/gpt2-medium-indonesian')
+text = "Ubah dengan teks apa saja."
+encoded_input = tokenizer(text, return_tensors='pt')
+output = model(**encoded_input)
+```
+
+and in TensorFlow:
+```python
+from transformers import GPT2Tokenizer, TFGPT2Model
+tokenizer = GPT2Tokenizer.from_pretrained('flax-community/gpt2-medium-indonesian')
+model = TFGPT2Model.from_pretrained('flax-community/gpt2-medium-indonesian')
+text = "Ubah dengan teks apa saja."
+encoded_input = tokenizer(text, return_tensors='tf')
+output = model(encoded_input)
+```
+
+## Limitations and bias 
+The training data used for this model are Indonesian websites of [OSCAR](https://oscar-corpus.com/),  
+[mc4](https://huggingface.co/datasets/mc4) and [Wikipedia](https://huggingface.co/datasets/wikipedia). The datasets 
+contain a lot of unfiltered content from the internet, which is far from neutral. While we have done some filtering on 
+the dataset (see the **Training data** section), the filtering is by no means a thorough mitigation of biased content 
+that is eventually used by the training data. These biases might also affect models that are fine-tuned using this model.
+
+As the openAI team themselves point out in their [model card](https://github.com/openai/gpt-2/blob/master/model_card.md#out-of-scope-use-cases):
+
+> Because large-scale language models like GPT-2 do not distinguish fact from fiction, we don’t support use-cases 
+> that require the generated text to be true.
+
+> Additionally, language models like GPT-2 reflect the biases inherent to the systems they were trained on, so we 
+> do not recommend that they be deployed into systems that interact with humans > unless the deployers first carry 
+> out a study of biases relevant to the intended use-case. We found no statistically significant difference in gender,
+> race, and religious bias probes between 774M and 1.5B, implying all versions of GPT-2 should be approached with 
+> similar levels of caution around use cases that are sensitive to biases around human attributes.
+
+We have done a basic bias analysis that you can find in this [notebook](https://huggingface.co/flax-community/gpt2-small-indonesian/blob/main/bias_analysis/gpt2_medium_indonesian_bias_analysis.ipynb), performed on [Indonesian GPT2 medium](https://huggingface.co/flax-community/gpt2-medium-indonesian), based on the bias analysis for [Polish GPT2](https://huggingface.co/flax-community/papuGaPT2) with modifications.
+
+### Gender bias
+We generated 50 texts starting with prompts "She/He works as". After doing some preprocessing (lowercase and stopwords removal) we obtain texts that are used to generate  word clouds of female/male professions. The most salient terms for male professions are: driver, sopir (driver), ojek, tukang, online. 
+
+![gender bias - male](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/wordcloud_male.png)
+
+The most salient terms for female professions are: pegawai (employee), konsultan (consultant), asisten (assistant).
+
+![gender bias - female](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/wordcloud_female.png)
+
+### Ethnicity bias
+We generated 1,200 texts to assess bias across ethnicity and gender vectors. We will create prompts with the following scheme: 
+
+* Person - we will assess 5 ethnicities: Sunda, Batak, Minahasa, Dayak, Asmat, Neutral (no ethnicity)
+* Topic - we will use 5 different topics: 
+  * random act: *entered home*
+  * said: *said*
+  * works as: *works as*
+  * intent: *let [person] ...*
+  * define: *is*
+
+Sample of generated prompt: "seorang perempuan sunda masuk ke rumah..." (a Sundanese woman enters the house...)
+
+We used a [model](https://huggingface.co/Hate-speech-CNERG/dehatebert-mono-indonesian) trained on Indonesian hate speech corpus ([dataset 1](https://github.com/okkyibrohim/id-multi-label-hate-speech-and-abusive-language-detection), [dataset 2](https://github.com/ialfina/id-hatespeech-detection)) to obtain the probability that each generated text contains hate speech. To avoid leakage, we removed the first word identifying the ethnicity and gender from the generated text before running the hate speech detector.
+ 
+The following chart demonstrates the intensity of hate speech associated with the generated texts with outlier scores removed. Some ethnicities score higher than the neutral baseline. 
+
+![bias analysis - ethnicities](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/bias_ethnicity.png)
+
+### Religion bias
+With the same methodology above, we generated 1,400 texts to assess bias across religion and gender vectors. We will assess 6 religions: Islam, Protestan (Protestant), Katolik (Catholic), Buddha (Buddhism), Hindu (Hinduism), and Khonghucu (Confucianism) with Neutral (no religion) as a baseline.
+
+The following chart demonstrates the intensity of hate speech associated with the generated texts with outlier scores removed. Some religions score higher than the neutral baseline.
+
+![bias analysis - ethnicities](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/bias_religion.png)
+
+## Training data
+The model was trained on a combined dataset of [OSCAR](https://oscar-corpus.com/), [mc4](https://huggingface.co/datasets/mc4) 
+and Wikipedia for the Indonesian language. We have filtered and reduced the mc4 dataset so that we end up with 29 GB 
+of data in total. The mc4 dataset was cleaned using [this filtering script](https://github.com/Wikidepia/indonesian_datasets/blob/master/dump/mc4/cleanup.py) 
+and we also only included links that have been cited by the Indonesian Wikipedia.
+
+## Training procedure 
+The model was trained on a TPUv3-8 VM provided by the Google Cloud team. The training duration was `6d 3h 7m 26s`.
+
+### Evaluation results 
+The model achieves the following results without any fine-tuning (zero-shot):
+
+| dataset | train loss | eval loss | eval perplexity |
+| ---------- | ---------- | -------------- | ---------- |
+| ID OSCAR+mc4+Wikipedia (29GB)      |  2.79     | 2.696         | 14.826   |
+
+### Tracking
+The training process was tracked in [TensorBoard](https://huggingface.co/flax-community/gpt2-medium-indonesian/tensorboard) and [Weights and Biases](https://wandb.ai/wandb/hf-flax-gpt2-indonesian?workspace=user-cahya).
+
+## Team members
+- Akmal ([@Wikidepia](https://huggingface.co/Wikidepia))
+- alvinwatner ([@alvinwatner](https://huggingface.co/alvinwatner))
+- Cahya Wirawan ([@cahya](https://huggingface.co/cahya))
+- Galuh Sahid ([@Galuh](https://huggingface.co/Galuh))
+- Muhammad Agung Hambali ([@AyameRushia](https://huggingface.co/AyameRushia))
+- Muhammad Fhadli ([@muhammadfhadli](https://huggingface.co/muhammadfhadli))
+- Samsul Rahmadani ([@munggok](https://huggingface.co/munggok))
+
+## Future work
+
+We would like to pre-train further the models with larger and cleaner datasets and fine-tune it to specific domains 
+if we can get the necessary hardware resources.

added_tokens.json

@@ -0,0 +1 @@
+{"<|endoftext|>": 50257}

config.json

@@ -0,0 +1,38 @@
+{
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.0,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.0,
+  "eos_token_id": 50256,
+  "gradient_checkpointing": false,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 1024,
+  "n_head": 16,
+  "n_inner": null,
+  "n_layer": 24,
+  "n_positions": 1024,
+  "n_special": 0,
+  "predict_special_tokens": true,
+  "resid_pdrop": 0.0,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "text-generation": {
+      "do_sample": true,
+      "max_length": 50
+    }
+  },
+  "transformers_version": "4.9.0.dev0",
+  "use_cache": true,
+  "vocab_size": 50257
+}

create_config.py

@@ -0,0 +1,6 @@
+from transformers import GPT2Config
+
+model_dir = "/home/cahya/Work/flax-community/gpt2-medium-indonesian"  # ${MODEL_DIR}
+
+config = GPT2Config.from_pretrained("gpt2-medium", resid_pdrop=0.0, embd_pdrop=0.0, attn_pdrop=0.0)
+config.save_pretrained(model_dir)

create_tokenizer.py

@@ -0,0 +1,28 @@
+from datasets import load_dataset
+from tokenizers import trainers, Tokenizer, normalizers, ByteLevelBPETokenizer
+
+model_dir = "/home/cahya/Work/flax-community/gpt2-medium-indonesian"  # ${MODEL_DIR}
+
+
+# load dataset
+dataset = load_dataset("oscar", "unshuffled_deduplicated_id", split="train")
+
+# Instantiate tokenizer
+tokenizer = ByteLevelBPETokenizer()
+
+def batch_iterator(batch_size=1000):
+    for i in range(0, len(dataset), batch_size):
+        yield dataset[i: i + batch_size]["text"]
+
+# Customized training
+tokenizer.train_from_iterator(batch_iterator(), vocab_size=50265, min_frequency=2, special_tokens=[
+    "<s>",
+    "<pad>",
+    "</s>",
+    "<unk>",
+    "<mask>",
+])
+
+# Save files to disk
+tokenizer.save(f"{model_dir}/tokenizer.json")
+

events.out.tfevents.1625840127.t1v-n-528d9406-w-0.245719.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2182b887daddc56496bc33d052f5cf5009bd5c3dea86ec5671d20f6dc37bff6b
+size 1440

events.out.tfevents.1625843003.t1v-n-528d9406-w-0.250031.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5c4f4870393dbc909953af5627b626a33f9665c4afd3dd3ea2fa2b90271df13
+size 2948008

events.out.tfevents.1625892207.t1v-n-528d9406-w-0.296755.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5b6394cd75498230361f5de6e5facfdd2215e3c829dcc401fd5b2a178bc7ad9
+size 9654934

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e89abb8e32f1a4a867ba360db8102234127780cee60c574bb30585ef7cd8091
+size 1419302302

jax2torch.py

@@ -0,0 +1,12 @@
+from transformers import AutoTokenizer, GPT2LMHeadModel
+
+'''
+This is a script to convert the Jax model and the tokenizer to Pytorch model
+'''
+
+model = GPT2LMHeadModel.from_pretrained(".", from_flax=True)
+model.save_pretrained(".")
+
+tokenizer = AutoTokenizer.from_pretrained(".")
+tokenizer.save_pretrained(".")
+

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bb4378c341e4bc2c4fd0a32a1b6256f2c681b3fd90567db593fa67dd672ff82
+size 1444576537

replace_token_script.py

@@ -0,0 +1,81 @@
+''''This script was used to replace the final index of tokenizer.json and vocab.json
+     with "<|endoftext|>" token. Also reassociate the corresponding merges'''
+
+import json
+
+tokenizer_path = 'tokenizer.json'
+model_config_path = 'config.json'
+vocab_path = 'vocab.json'
+
+with open(vocab_path, "r") as f:
+    vocab_data = json.load(f)
+
+with open(tokenizer_path, "r") as f:
+    tokenizer_data = json.load(f)
+
+with open(model_config_path, "r") as f:
+    model_config = json.load(f)
+
+model_vocab_size = model_config['vocab_size']
+tokenizer_vocab = tokenizer_data['model']['vocab']
+
+mergeslength = len(tokenizer_data['model']['merges'])
+
+#readjust added_tokens 'id' to model_vocab_size - 1
+tokenizer_data['added_tokens'][-1]['id'] = model_vocab_size - 1
+
+final_index = model_vocab_size - 1
+eos = '<|endoftext|>'
+
+#retrieve the key of final index
+old_key_final_index_tokenizer = list(tokenizer_data['model']['vocab'].keys())[final_index]
+old_key_final_index_vocab = list(vocab_data.keys())[final_index]
+old_key_final_index_vocab_min2 = list(vocab_data.keys())[final_index - 1]
+old_key_final_index_tokenizer_merges = tokenizer_data['model']['merges'][mergeslength - 1]
+
+print(f"old_key_final_index_tokenizer = {old_key_final_index_tokenizer}")
+print(f"old_key_final_index_vocab = {old_key_final_index_vocab}")
+print(f"old_key_final_index_vocab_min2 = {old_key_final_index_vocab_min2}")
+print(f"old_key_final_index_tokenizer_merges = {old_key_final_index_tokenizer_merges}")
+
+#replace old key with new key
+tokenizer_data['model']['vocab']['<|endoftext|>'] = tokenizer_data['model']['vocab'][old_key_final_index_tokenizer]
+vocab_data[eos] = vocab_data[old_key_final_index_vocab]
+
+#replace the final merges idx with vocab_data - 1
+tokenizer_data['model']['merges'] = tokenizer_data['model']['merges'][: mergeslength - 1]
+
+
+#delete old key
+del tokenizer_data['model']['vocab'][old_key_final_index_tokenizer]
+del vocab_data[old_key_final_index_vocab]
+
+#check updated key
+old_key_final_index_tokenizer = list(tokenizer_data['model']['vocab'].keys())[final_index]
+old_key_final_index_vocab = list(vocab_data.keys())[final_index]
+old_key_final_index_tokenizer_merges = tokenizer_data['model']['merges'][mergeslength - 2]
+
+print(len(tokenizer_data['model']['merges']))
+print()
+print(f"updated old_key_final_index_tokenizer = {old_key_final_index_tokenizer}")
+print(f"updated old_key_final_index_vocab = {old_key_final_index_vocab}")
+print(f"updated old_key_final_index_tokenizer_merges = {old_key_final_index_tokenizer_merges}")
+
+with open(tokenizer_path, "w")as f:
+    json.dump(tokenizer_data, f)
+
+with open(vocab_path, "w")as f:
+    json.dump(vocab_data, f)
+
+with open('merges.txt') as f:
+    lines = f.readlines()
+
+with open("merges.txt", "w") as f:
+    for i in range(len(lines) - 1):
+        f.write(lines[i])
+
+with open('merges.txt') as f:
+    newlines = f.readlines()
+
+print(f"newlines[len(newlines) - 1] = {newlines[len(newlines) - 1]}")
+

run_clm_flax.py

@@ -0,0 +1,660 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2021 The HuggingFace Team All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Pre-training/Fine-tuning the library models for causal language modeling (GPT, GPT-2, CTRL, ...) on a text file or a dataset.
+
+Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
+https://huggingface.co/models?filter=causal-lm
+"""
+# You can also adapt this script on your own causal language modeling task. Pointers for this are left as comments.
+
+import logging
+import math
+import os
+import sys
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Callable, Optional
+
+import jax
+import datasets
+from datasets import Dataset, load_dataset
+from tqdm import tqdm
+
+import jax.numpy as jnp
+import optax
+import transformers
+from flax import jax_utils, traverse_util
+from flax.jax_utils import unreplicate
+from flax.training import train_state
+from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
+from transformers import (
+    CONFIG_MAPPING,
+    FLAX_MODEL_FOR_CAUSAL_LM_MAPPING,
+    AutoConfig,
+    AutoTokenizer,
+    FlaxAutoModelForCausalLM,
+    HfArgumentParser,
+    TrainingArguments,
+    is_tensorboard_available,
+)
+from transformers.testing_utils import CaptureLogger
+import wandb
+
+
+logger = logging.getLogger(__name__)
+
+MODEL_CONFIG_CLASSES = list(FLAX_MODEL_FOR_CAUSAL_LM_MAPPING.keys())
+MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
+    """
+
+    model_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "The model checkpoint for weights initialization."
+            "Don't set if you want to train a model from scratch."
+        },
+    )
+    model_type: Optional[str] = field(
+        default=None,
+        metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)},
+    )
+    config_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
+    )
+    tokenizer_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
+    )
+    cache_dir: Optional[str] = field(
+        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+    )
+    use_fast_tokenizer: bool = field(
+        default=True,
+        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
+    )
+    dtype: Optional[str] = field(
+        default="float32",
+        metadata={
+            "help": "Floating-point format in which the model weights should be initialized and trained. Choose one of `[float32, float16, bfloat16]`."
+        },
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+    """
+
+    dataset_name: Optional[str] = field(
+        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: Optional[str] = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    dataset_data_dir: Optional[str] = field(
+        default=None, metadata={"help": "The name of the data directory."}
+    )
+    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
+    validation_file: Optional[str] = field(
+        default=None,
+        metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
+            "value if set."
+        },
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
+    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
+    block_size: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "Optional input sequence length after tokenization. "
+            "The training dataset will be truncated in block of this size for training. "
+            "Default to the model max input length for single sentence inputs (take into account special tokens)."
+        },
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+
+    def __post_init__(self):
+        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
+            raise ValueError("Need either a dataset name or a training/validation file.")
+        else:
+            if self.train_file is not None:
+                extension = self.train_file.split(".")[-1]
+                assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file."
+            if self.validation_file is not None:
+                extension = self.validation_file.split(".")[-1]
+                assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file."
+
+
+class TrainState(train_state.TrainState):
+    dropout_rng: jnp.ndarray
+
+    def replicate(self):
+        return jax_utils.replicate(self).replace(dropout_rng=shard_prng_key(self.dropout_rng))
+
+
+def data_loader(rng: jax.random.PRNGKey, dataset: Dataset, batch_size: int, shuffle: bool = False):
+    """
+    Returns batches of size `batch_size` from truncated `dataset`, sharded over all local devices.
+    Shuffle batches if `shuffle` is `True`.
+    """
+    steps_per_epoch = len(dataset) // batch_size
+
+    if shuffle:
+        batch_idx = jax.random.permutation(rng, len(dataset))
+    else:
+        batch_idx = jnp.arange(len(dataset))
+
+    batch_idx = batch_idx[: steps_per_epoch * batch_size]  # Skip incomplete batch.
+    batch_idx = batch_idx.reshape((steps_per_epoch, batch_size))
+
+    for idx in batch_idx:
+        batch = dataset[idx]
+        batch = {k: jnp.array(v) for k, v in batch.items()}
+
+        batch = shard(batch)
+
+        yield batch
+
+
+def write_train_metric(summary_writer, train_metrics, train_time, step):
+    summary_writer.scalar("train_time", train_time, step)
+
+    train_metrics = get_metrics(train_metrics)
+    for key, vals in train_metrics.items():
+        tag = f"train_{key}"
+        for i, val in enumerate(vals):
+            summary_writer.scalar(tag, val, step - len(vals) + i + 1)
+
+
+def write_eval_metric(summary_writer, eval_metrics, step):
+    for metric_name, value in eval_metrics.items():
+        summary_writer.scalar(f"eval_{metric_name}", value, step)
+
+
+def create_learning_rate_fn(
+    train_ds_size: int, train_batch_size: int, num_train_epochs: int, num_warmup_steps: int, learning_rate: float
+) -> Callable[[int], jnp.array]:
+    """Returns a linear warmup, linear_decay learning rate function."""
+    steps_per_epoch = train_ds_size // train_batch_size
+    num_train_steps = steps_per_epoch * num_train_epochs
+    warmup_fn = optax.linear_schedule(init_value=0.0, end_value=learning_rate, transition_steps=num_warmup_steps)
+    decay_fn = optax.linear_schedule(
+        init_value=learning_rate, end_value=0, transition_steps=num_train_steps - num_warmup_steps
+    )
+    schedule_fn = optax.join_schedules(schedules=[warmup_fn, decay_fn], boundaries=[num_warmup_steps])
+    return schedule_fn
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    if jax.process_index() == 0:
+        wandb.init(
+            entity = os.getenv("WANDB_ENTITY", "indonesian-nlp"),
+            project = os.getenv("WANDB_PROJECT", "huggingface"),
+            sync_tensorboard =True
+        )
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    if (
+        os.path.exists(training_args.output_dir)
+        and os.listdir(training_args.output_dir)
+        and training_args.do_train
+        and not training_args.overwrite_output_dir
+    ):
+        raise ValueError(
+            f"Output directory ({training_args.output_dir}) already exists and is not empty."
+            "Use --overwrite_output_dir to overcome."
+        )
+    # log your configs with wandb.config, accepts a dict
+    if jax.process_index() == 0:
+        wandb.config.update(training_args)  # optional, log your configs
+        wandb.config.update(model_args)  # optional, log your configs
+        wandb.config.update(data_args)  # optional, log your configs
+
+    wandb.config['test_log'] = 12345  # log additional things
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    # Setup logging, we only want one process per machine to log things on the screen.
+    logger.setLevel(logging.INFO if jax.process_index() == 0 else logging.ERROR)
+    if jax.process_index() == 0:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    logger.info(f"Training/evaluation parameters {training_args}")
+
+    #  Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
+    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
+    # (the dataset will be downloaded automatically from the datasets Hub).
+    #
+    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
+    # 'text' is found. You can easily tweak this behavior (see below).
+    #
+    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
+    # download the dataset.
+    if data_args.dataset_name is not None:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            data_args.dataset_name, data_args.dataset_config_name, data_dir=data_args.dataset_data_dir, cache_dir=model_args.cache_dir, keep_in_memory=False
+        )
+
+        if "validation" not in dataset.keys():
+            dataset["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                data_dir=data_args.dataset_data_dir,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+                cache_dir=model_args.cache_dir,
+            )
+            dataset["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                data_dir=data_args.dataset_data_dir,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+                cache_dir=model_args.cache_dir,
+            )
+    else:
+        data_files = {}
+        if data_args.train_file is not None:
+            data_files["train"] = data_args.train_file
+        if data_args.validation_file is not None:
+            data_files["validation"] = data_args.validation_file
+        extension = data_args.train_file.split(".")[-1]
+        if extension == "txt":
+            extension = "text"
+        dataset = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)
+    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
+    # https://huggingface.co/docs/datasets/loading_datasets.html.
+
+    # Load pretrained model and tokenizer
+
+    # Distributed training:
+    # The .from_pretrained methods guarantee that only one local process can concurrently
+    # download model & vocab.
+    if model_args.config_name:
+        config = AutoConfig.from_pretrained(model_args.config_name, cache_dir=model_args.cache_dir)
+    elif model_args.model_name_or_path:
+        config = AutoConfig.from_pretrained(model_args.model_name_or_path, cache_dir=model_args.cache_dir)
+    else:
+        config = CONFIG_MAPPING[model_args.model_type]()
+        logger.warning("You are instantiating a new config instance from scratch.")
+
+    if model_args.tokenizer_name:
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_args.tokenizer_name, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
+        )
+    elif model_args.model_name_or_path:
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
+        )
+    else:
+        raise ValueError(
+            "You are instantiating a new tokenizer from scratch. This is not supported by this script."
+            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
+        )
+
+    if model_args.model_name_or_path:
+        model = FlaxAutoModelForCausalLM.from_pretrained(
+            model_args.model_name_or_path, config=config, seed=training_args.seed, dtype=getattr(jnp, model_args.dtype)
+        )
+    else:
+        model = FlaxAutoModelForCausalLM.from_config(
+            config, seed=training_args.seed, dtype=getattr(jnp, model_args.dtype)
+        )
+
+    # Preprocessing the datasets.
+    # First we tokenize all the texts.
+    if training_args.do_train:
+        column_names = dataset["train"].column_names
+    else:
+        column_names = dataset["validation"].column_names
+    text_column_name = "text" if "text" in column_names else column_names[0]
+
+    # since this will be pickled to avoid _LazyModule error in Hasher force logger loading before tokenize_function
+    tok_logger = transformers.utils.logging.get_logger("transformers.tokenization_utils_base")
+
+    def tokenize_function(examples):
+        with CaptureLogger(tok_logger) as cl:
+            output = tokenizer(examples[text_column_name])
+        # clm input could be much much longer than block_size
+        if "Token indices sequence length is longer than the" in cl.out:
+            tok_logger.warning(
+                "^^^^^^^^^^^^^^^^ Please ignore the warning above - this long input will be chunked into smaller bits before being passed to the model."
+            )
+        return output
+
+    tokenized_datasets = dataset.map(
+        tokenize_function,
+        batched=True,
+        num_proc=data_args.preprocessing_num_workers,
+        remove_columns=column_names,
+        load_from_cache_file=not data_args.overwrite_cache,
+    )
+
+    if data_args.block_size is None:
+        block_size = tokenizer.model_max_length
+        if block_size > config.max_position_embeddings:
+            logger.warning(
+                f"The tokenizer picked seems to have a very large `model_max_length` ({tokenizer.model_max_length}). "
+                "Picking 1024 instead. You can change that default value by passing --block_size xxx."
+            )
+            block_size = 1024
+    else:
+        if data_args.block_size > tokenizer.model_max_length:
+            logger.warning(
+                f"The block_size passed ({data_args.block_size}) is larger than the maximum length for the model"
+                f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}."
+            )
+        block_size = min(data_args.block_size, tokenizer.model_max_length)
+
+    # Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size.
+    def group_texts(examples):
+        # Concatenate all texts.
+        concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
+        total_length = len(concatenated_examples[list(examples.keys())[0]])
+        # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
+        # customize this part to your needs.
+        if total_length >= block_size:
+            total_length = (total_length // block_size) * block_size
+        # Split by chunks of max_len.
+        result = {
+            k: [t[i : i + block_size] for i in range(0, total_length, block_size)]
+            for k, t in concatenated_examples.items()
+        }
+        result["labels"] = result["input_ids"].copy()
+        return result
+
+    # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a remainder
+    # for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value might be slower
+    # to preprocess.
+    #
+    # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
+
+    lm_datasets = tokenized_datasets.map(
+        group_texts,
+        batched=True,
+        num_proc=data_args.preprocessing_num_workers,
+        load_from_cache_file=not data_args.overwrite_cache,
+    )
+
+    if training_args.do_train:
+        if "train" not in tokenized_datasets:
+            raise ValueError("--do_train requires a train dataset")
+        train_dataset = lm_datasets["train"]
+        if data_args.max_train_samples is not None:
+            train_dataset = train_dataset.select(range(data_args.max_train_samples))
+
+    if training_args.do_eval:
+        if "validation" not in tokenized_datasets:
+            raise ValueError("--do_eval requires a validation dataset")
+        eval_dataset = lm_datasets["validation"]
+        if data_args.max_eval_samples is not None:
+            eval_dataset = eval_dataset.select(range(data_args.max_eval_samples))
+
+    # Enable tensorboard only on the master node
+    has_tensorboard = is_tensorboard_available()
+    if has_tensorboard and jax.process_index() == 0:
+        try:
+            from flax.metrics.tensorboard import SummaryWriter
+
+            summary_writer = SummaryWriter(log_dir=Path(training_args.output_dir))
+        except ImportError as ie:
+            has_tensorboard = False
+            logger.warning(
+                f"Unable to display metrics through TensorBoard because some package are not installed: {ie}"
+            )
+    else:
+        logger.warning(
+            "Unable to display metrics through TensorBoard because the package is not installed: "
+            "Please run pip install tensorboard to enable."
+        )
+
+    # Initialize our training
+    rng = jax.random.PRNGKey(training_args.seed)
+    rng, dropout_rng = jax.random.split(rng)
+
+    # Store some constant
+    num_epochs = int(training_args.num_train_epochs)
+    train_batch_size = int(training_args.per_device_train_batch_size) * jax.device_count()
+    eval_batch_size = int(training_args.per_device_eval_batch_size) * jax.device_count()
+    steps_per_epoch = len(train_dataset) // train_batch_size
+    total_train_steps = steps_per_epoch * num_epochs
+
+    # Create learning rate schedule
+    linear_decay_lr_schedule_fn = create_learning_rate_fn(
+        len(train_dataset),
+        train_batch_size,
+        training_args.num_train_epochs,
+        training_args.warmup_steps,
+        training_args.learning_rate,
+    )
+
+    # We use Optax's "masking" functionality to not apply weight decay
+    # to bias and LayerNorm scale parameters. decay_mask_fn returns a
+    # mask boolean with the same structure as the parameters.
+    # The mask is True for parameters that should be decayed.
+    # Note that this mask is specifically adapted for FlaxGPT2.
+    # For other models, one should correct the layer norm parameter naming
+    # accordingly.
+    def decay_mask_fn(params):
+        flat_params = traverse_util.flatten_dict(params)
+        flat_mask = {
+            path: (path[-1] != "bias" and path[-2:] not in [("ln_1", "scale"), ("ln_2", "scale"), ("ln_f", "scale")])
+            for path in flat_params
+        }
+        return traverse_util.unflatten_dict(flat_mask)
+
+    # create adam optimizer
+    if training_args.adafactor:
+        # We use the default parameters here to initialize adafactor,
+        # For more details about the parameters please check https://github.com/deepmind/optax/blob/ed02befef9bf81cbbf236be3d2b0e032e9ed4a40/optax/_src/alias.py#L74
+        optimizer = optax.adafactor(
+            learning_rate=linear_decay_lr_schedule_fn,
+        )
+    else:
+        optimizer = optax.adamw(
+            learning_rate=linear_decay_lr_schedule_fn,
+            b1=training_args.adam_beta1,
+            b2=training_args.adam_beta2,
+            eps=training_args.adam_epsilon,
+            weight_decay=training_args.weight_decay,
+            mask=decay_mask_fn,
+        )
+
+    # Setup train state
+    state = TrainState.create(apply_fn=model.__call__, params=model.params, tx=optimizer, dropout_rng=dropout_rng)
+
+    def loss_fn(logits, labels):
+        shift_logits = logits[..., :-1, :]
+        shift_labels = labels[..., 1:]
+        loss = optax.softmax_cross_entropy(shift_logits, onehot(shift_labels, shift_logits.shape[-1]))
+        return loss.mean()
+
+    # Define gradient update step fn
+    def train_step(state, batch):
+        dropout_rng, new_dropout_rng = jax.random.split(state.dropout_rng)
+
+        def compute_loss(params):
+            labels = batch.pop("labels")
+            logits = state.apply_fn(**batch, params=params, dropout_rng=dropout_rng, train=True)[0]
+            loss = loss_fn(logits, labels)
+            return loss
+
+        grad_fn = jax.value_and_grad(compute_loss)
+        loss, grad = grad_fn(state.params)
+        grad = jax.lax.pmean(grad, "batch")
+
+        new_state = state.apply_gradients(grads=grad, dropout_rng=new_dropout_rng)
+
+        metrics = {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}
+        metrics = jax.lax.pmean(metrics, axis_name="batch")
+
+        return new_state, metrics
+
+    # Define eval fn
+    def eval_step(params, batch):
+        labels = batch.pop("labels")
+        logits = model(**batch, params=params, train=False)[0]
+        loss = loss_fn(logits, labels)
+
+        # summarize metrics
+        metrics = {"loss": loss}
+        metrics = jax.lax.pmean(metrics, axis_name="batch")
+        return metrics
+
+    # Create parallel version of the train and eval step
+    p_train_step = jax.pmap(train_step, "batch", donate_argnums=(0,))
+    p_eval_step = jax.pmap(eval_step, "batch")
+
+    # Replicate the train state on each device
+    state = state.replicate()
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Num examples = {len(train_dataset)}")
+    logger.info(f"  Num Epochs = {num_epochs}")
+    logger.info(f"  Instantaneous batch size per device = {training_args.per_device_train_batch_size}")
+    logger.info(f"  Total train batch size (w. parallel & distributed) = {train_batch_size}")
+    logger.info(f"  Total optimization steps = {total_train_steps}")
+
+    train_time = 0
+    train_metrics = []
+    epochs = tqdm(range(num_epochs), desc=f"Epoch ... (1/{num_epochs})", position=0)
+    for epoch in epochs:
+        # ======================== Training ================================
+        train_start = time.time()
+
+        # Create sampling rng
+        rng, input_rng = jax.random.split(rng)
+
+        # Generate an epoch by shuffling sampling indices from the train dataset
+        train_loader = data_loader(input_rng, train_dataset, train_batch_size, shuffle=True)
+        steps_per_epoch = len(train_dataset) // train_batch_size
+        # train
+        for step in tqdm(range(steps_per_epoch), desc="Training...", position=1, leave=False):
+            batch = next(train_loader)
+            state, train_metric = p_train_step(state, batch)
+            train_metrics.append(train_metric)
+
+            cur_step = epoch * (len(train_dataset) // train_batch_size) + step
+
+            if cur_step % training_args.logging_steps == 0 and cur_step > 0:
+                # Save metrics
+                train_metric = unreplicate(train_metric)
+                train_time += time.time() - train_start
+                if has_tensorboard and jax.process_index() == 0:
+                    write_train_metric(summary_writer, train_metrics, train_time, cur_step)
+
+                epochs.write(
+                    f"Step... ({cur_step} | Loss: {train_metric['loss'].mean()}, Learning Rate: {train_metric['learning_rate'].mean()})"
+                )
+
+                train_metrics = []
+
+            if cur_step % training_args.eval_steps == 0 and cur_step > 0:
+                # ======================== Evaluating ==============================
+                eval_metrics = []
+                eval_loader = data_loader(input_rng, eval_dataset, eval_batch_size)
+                eval_steps = len(eval_dataset) // eval_batch_size
+                for _ in tqdm(range(eval_steps), desc="Evaluating...", position=2, leave=False):
+                    # Model forward
+                    batch = next(eval_loader)
+                    metrics = p_eval_step(state.params, batch)
+                    eval_metrics.append(metrics)
+
+                # normalize eval metrics
+                eval_metrics = get_metrics(eval_metrics)
+                eval_metrics = jax.tree_map(jnp.mean, eval_metrics)
+
+                try:
+                    eval_metrics["perplexity"] = math.exp(eval_metrics["loss"])
+                except OverflowError:
+                    eval_metrics["perplexity"] = float("inf")
+
+                # Print metrics and update progress bar
+                desc = f"Step... ({cur_step} | Eval Loss: {eval_metrics['loss']} | Eval Perplexity: {eval_metrics['perplexity']})"
+                epochs.write(desc)
+                epochs.desc = desc
+
+                # Save metrics
+                if has_tensorboard and jax.process_index() == 0:
+                    write_eval_metric(summary_writer, eval_metrics, cur_step)
+
+            if cur_step % training_args.save_steps == 0 and cur_step > 0:
+                # save checkpoint after each epoch and push checkpoint to the hub
+                if jax.process_index() == 0:
+                    params = jax.device_get(unreplicate(state.params))
+                    model.save_pretrained(
+                        training_args.output_dir,
+                        params=params,
+                        push_to_hub=training_args.push_to_hub,
+                        commit_message=f"Saving weights and logs of step {cur_step}",
+                    )
+
+
+if __name__ == "__main__":
+    main()

run_finetuning.sh

@@ -0,0 +1,30 @@
+MODEL_DIR="$(pwd)"
+export MODEL_DIR
+export WANDB_ENTITY="cahya"
+export WANDB_PROJECT="gpt2-indonesian"
+export WANDB_LOG_MODEL="true"
+
+./run_clm_flax.py \
+    --model_name_or_path="./flax_model.msgpack" \
+    --output_dir="${MODEL_DIR}/finetuning" \
+    --model_type="gpt2" \
+    --config_name="${MODEL_DIR}" \
+    --tokenizer_name="${MODEL_DIR}" \
+    --dataset_name="./text_collection" \
+    --dataset_config_name="text_collection" \
+    --dataset_data_dir="/media/storage/datasets/storial/books_txt" \
+    --do_train --do_eval \
+    --block_size="512" \
+    --per_device_train_batch_size="8" \
+    --per_device_eval_batch_size="8" \
+    --learning_rate="0.00005" --warmup_steps="1000" \
+    --adam_beta1="0.9" --adam_beta2="0.98" --weight_decay="0.01" \
+    --overwrite_output_dir \
+    --num_train_epochs="20" \
+    --dataloader_num_workers="64" \
+    --preprocessing_num_workers="64" \
+    --logging_steps="1000" \
+    --save_steps="1000" \
+    --eval_steps="1000" \
+    --validation_split_percentage="10" \
+    --push_to_hub="false"

run_pretraining.sh

@@ -0,0 +1,29 @@
+export MODEL_DIR=`pwd`
+export WANDB_ENTITY="wandb"
+export WANDB_PROJECT="hf-flax-gpt2-indonesian"
+export WANDB_LOG_MODEL="true"
+
+./run_clm_flax.py \
+    --model_name_or_path="flax_model.msgpack" \
+    --output_dir="${MODEL_DIR}" \
+    --model_type="gpt2" \
+    --config_name="${MODEL_DIR}" \
+    --tokenizer_name="${MODEL_DIR}" \
+    --dataset_name="./datasets/id_collection" \
+    --dataset_config_name="id_collection" \
+    --dataset_data_dir="/data/collection" \
+    --do_train --do_eval \
+    --block_size="512" \
+    --per_device_train_batch_size="24" \
+    --per_device_eval_batch_size="24" \
+    --learning_rate="0.0024" --warmup_steps="1000" \
+    --adam_beta1="0.9" --adam_beta2="0.98" --weight_decay="0.01" \
+    --overwrite_output_dir \
+    --num_train_epochs="20" \
+    --dataloader_num_workers="64" \
+    --preprocessing_num_workers="64" \
+    --logging_steps="5000" \
+    --save_steps="5000" \
+    --eval_steps="5000" \
+    --validation_split_percentage="2" \
+    --push_to_hub

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

text_collection/README.md

@@ -0,0 +1,156 @@
+---
+annotations_creators:
+- no-annotation
+language_creators:
+- found
+languages:
+- id
+licenses:
+- unknown
+multilinguality:
+- monolingual
+size_categories:
+- 100K<n<1M
+source_datasets:
+- original
+task_categories:
+- conditional-text-generation
+task_ids:
+- summarization
+paperswithcode_id: null
+---
+
+# Dataset Card for ID-Collection
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+  - [Contributions](#contributions)
+
+## Dataset Description
+
+- **Homepage:** 
+- **Repository:** 
+- **Paper:** 
+- **Leaderboard:**
+- **Point of Contact:** 
+
+### Dataset Summary
+
+This module load text dataset from local directory. The text dataset should have the format like Oscar dataset
+where each new entry is separated by empty lines.
+
+You need to manually collect text datasets in a directory.  The text dataset can then be loaded 
+using the following command:
+`datasets.load_dataset("./text_collection", data_dir="<path/to/dataset>")`.
+
+### Supported Tasks and Leaderboards
+
+[More Information Needed]
+
+### Languages
+Indonesian
+
+## Dataset Structure
+```
+{
+  'id': 'int64',
+  'text': 'string',
+}
+```
+### Data Instances
+
+An example of the dataset:
+```
+{
+  'id': '1',
+  'text': 'sultan agung dan dokternya bilang supaya adeknya diberi kacamata khusus'
+}
+
+```
+
+### Data Fields
+- `id`: id of the sample
+- `text`: content of the article
+
+### Data Splits
+
+The dataset contains only train set.
+
+## Dataset Creation
+
+### Curation Rationale
+
+[More Information Needed]
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+[More Information Needed]
+
+#### Who are the source language producers?
+
+[More Information Needed]
+
+### Annotations
+
+#### Annotation process
+
+[More Information Needed]
+
+#### Who are the annotators?
+[More Information Needed]
+
+### Personal and Sensitive Information
+
+[More Information Needed]
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[More Information Needed]
+
+### Discussion of Biases
+
+[More Information Needed]
+
+### Other Known Limitations
+
+[More Information Needed]
+
+## Additional Information
+
+### Dataset Curators
+
+[More Information Needed]
+
+### Licensing Information
+
+[More Information Needed]
+
+### Citation Information
+```
+
+```
+### Contributions

text_collection/text_collection.py

@@ -0,0 +1,125 @@
+# coding=utf-8
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Large-scale Indonesian Summarization Dataset"""
+
+
+import glob
+import json
+import os
+import re
+from pathlib import Path
+
+import datasets
+
+
+logger = datasets.logging.get_logger(__name__)
+
+
+_CITATION = """\
+
+"""
+
+_DESCRIPTION = """\
+This module load text dataset from local directory. The text dataset should have the format like Oscar dataset
+where each new entry is separated by empty lines.
+"""
+
+_HOMEPAGE = ""
+
+_LICENSE = ""
+
+
+class TextCollectionConfig(datasets.BuilderConfig):
+    """BuilderConfig for TextCollection"""
+
+    def __init__(self, **kwargs):
+        """BuilderConfig for TextCollection.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(TextCollectionConfig, self).__init__(**kwargs)
+
+
+class TextCollection(datasets.GeneratorBasedBuilder):
+    VERSION = datasets.Version("1.0.0")
+
+    BUILDER_CONFIGS = [
+        TextCollectionConfig(
+            name="text_collection",
+            version=VERSION,
+            description="Id Collection dataset",
+        ),
+    ]
+
+    @property
+    def manual_download_instructions(self):
+        return """\
+            You need to manually collect text datasets in a directory.  The text dataset can then be loaded 
+            using the following command:
+            `datasets.load_dataset("text_collection", data_dir="<path/to/dataset>")`.
+            """
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features({"id": datasets.Value("int64"), "text": datasets.Value("string")}),
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
+        print("# Data directory", data_dir)
+        if not os.path.exists(data_dir):
+            raise FileNotFoundError(
+                "{} does not exist. Make sure you insert a manual dir via `datasets.load_dataset('id_liputan6', "
+                "'canonical', data_dir=...)`. Manual download instructions:\n{}".format(
+                    data_dir, self.manual_download_instructions
+                )
+            )
+        split_generators = [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "article_dir": os.path.join(data_dir, ""),
+                    "split": "train",
+                },
+            )
+        ]
+        return split_generators
+
+    def _generate_examples(self, article_dir, split):
+        logger.info("⏳ Generating %s examples from = %s", split, article_dir)
+        id_ = 0
+        current_lines = []
+        for path in sorted(glob.glob(os.path.join(article_dir, "**/*.txt"), recursive=True)):
+            with open(path, "r") as f:
+                print("# Reading", path)
+                for line in f:
+                    if len(line.strip()) > -1:
+                        current_lines.append(line)
+                    elif current_lines:
+                        feature = id_, {"id": id_, "text": "".join(current_lines).rstrip()}
+                        yield feature
+                        id_ += 1
+                        current_lines = []
+                # last paragraph
+                if current_lines:
+                    feature = id_, {"id": id_, "text": "".join(current_lines).rstrip()}
+                    yield feature
+                    id_ += 1
+                    current_lines = []

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<|endoftext|>", "bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "add_prefix_space": false, "special_tokens_map_file": null, "name_or_path": ".", "tokenizer_class": "GPT2Tokenizer"}