File size: 6,420 Bytes
39a7504
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
from tokenizers import Tokenizer
from tokenizers.models import BPE
from tokenizers.trainers import BpeTrainer
from tokenizers.pre_tokenizers import (
    WhitespaceSplit,
    Punctuation,
    Sequence as PreSequence,
)
from tokenizers.normalizers import NFD, Lowercase, Sequence, StripAccents
from pathlib import Path
import yaml
from datasets import load_dataset
from dataset import English2HindiDataset, English2HindiDatasetTest
from torch.utils.data import DataLoader
import json
import random
from sklearn.model_selection import train_test_split


def load_config(config_path):
    with open(config_path, "r") as file:
        return yaml.safe_load(file)


def get_all_sentences(ds, lang):
    for item in ds:
        yield item[lang]


def get_or_build_tokenizer(tokenizer_path, ds, lang):
    tokenizer_path = Path(tokenizer_path)
    if not Path.exists(tokenizer_path):
        tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
        if lang == "en_text":
            tokenizer.normalizer = Sequence(
                [
                    NFD(),
                    StripAccents(),
                    Lowercase(),
                ]
            )
        else:
            tokenizer.normalizer = Sequence(
                [
                    NFD(),
                    Lowercase(),
                ]
            )

        tokenizer.pre_tokenizer = PreSequence([WhitespaceSplit(), Punctuation()])
        trainer = BpeTrainer(
            special_tokens=["[UNK]", "[PAD]", "[SOS]", "[EOS]"],
            min_frequency=3,
            vocab_size=60000,
        )
        tokenizer.train_from_iterator(get_all_sentences(ds, lang), trainer=trainer)
        tokenizer.save(str(tokenizer_path))
    else:
        tokenizer = Tokenizer.from_file(str(tokenizer_path))
    return tokenizer


def preprocess_to_json(dataset_hf, output_json_path, word_limit=77, char_limit=300):
    filtered_data = []

    for row in dataset_hf:
        if not row.get("translated", False):
            continue

        en_value = None
        en_conversations = row.get("en_conversations", [])
        for conv in en_conversations:
            if "human" in conv["from"]:
                en_value = conv["value"]
                break
        if en_value is None and len(en_conversations) > 0:
            en_value = en_conversations[0]["value"]

        hi_value = None
        hi_conversations = row.get("conversations", [])
        for conv in hi_conversations:
            if "human" in conv["from"]:
                hi_value = conv["value"]
                break
        if hi_value is None and len(hi_conversations) > 0:
            hi_value = hi_conversations[0]["value"]

        if en_value and hi_value:
            if (
                len(en_value.split()) > word_limit
                or len(en_value) > char_limit
                or len(hi_value.split()) > word_limit
                or len(hi_value) > char_limit
            ):
                continue

            filtered_data.append({"en_text": en_value, "hi_text": hi_value})

    with open(output_json_path, "w", encoding="utf-8") as f:
        json.dump(filtered_data, f, ensure_ascii=False, indent=2)



def create_resources():
    config_path = "config.yaml"
    config = load_config(config_path)

    dataset_json_path = config.get("dataset_path", "data/english2hindi_data.json")

    if not Path(dataset_json_path).exists():
        print(f"Dataset file {dataset_json_path} not found. Creating it...")
        dataset_hf = load_dataset("BhabhaAI/openhermes-2.5-hindi", split="train")
        preprocess_to_json(dataset_hf, dataset_json_path)
    else:
        print(f"Dataset file {dataset_json_path} already exists. Skipping preprocessing.")


    with open(dataset_json_path, "r", encoding="utf-8") as f:
        raw_data = json.load(f)



    tokenizer_src = get_or_build_tokenizer(
        config["src_tokenizer_file"], raw_data, config["src_lang"]
    )
    tokenizer_tgt = get_or_build_tokenizer(
        config["tgt_tokenizer_file"], raw_data, config["tgt_lang"]
    )


    test_pt_dataset = English2HindiDatasetTest(config["dataset_path"])

    print(len(test_pt_dataset))

    max_len_src = 0
    max_len_tgt = 0


    seq_len = config["seq_len"]

    if seq_len == 0:
        print("seq_len is 0, starting process...")

        
        for item in raw_data:
            src_ids = tokenizer_src.encode(item[config['src_lang']]).ids
            tgt_ids = tokenizer_tgt.encode(item[config['tgt_lang']]).ids
            max_len_src = max(max_len_src, len(src_ids))
            
            max_len_tgt = max(max_len_tgt, len(tgt_ids))


        print(max_len_src,max_len_tgt )  

        final_max_len = max(max_len_src, max_len_tgt) + 30

        config['seq_len'] = final_max_len


        with open("config.yaml", 'w') as f:
            yaml.safe_dump(config, f, default_flow_style=False)

        print(f'Updated seq_len to {final_max_len}')

    else:
        print("seq_len is not 0, skipping process.")

    random.seed(42)

    train_data, temp_data = train_test_split(raw_data, test_size=0.2, random_state=42)
    test_data, valid_data = train_test_split(temp_data, test_size=0.5, random_state=42)


    print("######################################################")

    train_dataset = English2HindiDataset(
        train_data,
        tokenizer_src,
        tokenizer_tgt,
        config["src_lang"],
        config["tgt_lang"],
        config["seq_len"],
    )

    valid_dataset = English2HindiDataset(
        valid_data,
        tokenizer_src,
        tokenizer_tgt,
        config["src_lang"],
        config["tgt_lang"],
        config["seq_len"],
    )

    test_dataset = English2HindiDataset(
        test_data,
        tokenizer_src,
        tokenizer_tgt,
        config["src_lang"],
        config["tgt_lang"],
        config["seq_len"],
    )


    train_dataloader = DataLoader(train_dataset, batch_size=config["batch_size"], shuffle=False)
    valid_dataloader = DataLoader(valid_dataset, batch_size=config["batch_size"],shuffle=True)
    test_dataloader = DataLoader(test_dataset, batch_size=config["batch_size"],shuffle=True)

    return train_dataloader,valid_dataloader,test_dataloader,tokenizer_src,tokenizer_tgt