inference_fertility.py

from datasets import concatenate_datasets, load_dataset, load_from_disk
import argparse
from tokenizers import Tokenizer, decoders, models, pre_tokenizers, processors, trainers
from transformers import GPT2TokenizerFast, AutoTokenizer
from datasets import config
import logging
from datasets import DatasetDict, Dataset
import csv
import time
import json
tokenizer = AutoTokenizer.from_pretrained('Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Base')
def initialize_logger(log_file):
    logging.basicConfig(filename=log_file, level=logging.INFO, format='%(asctime)s: %(message)s')

def log_parameters(vocab_size, pretrained_model, en_fertility_score, hi_fertility_score , ta_fertility_score , log_file='parameters.log'):
    initialize_logger(log_file)
    logging.info(f"Vocabulary Size: {vocab_size}, Tokenizer type: {pretrained_model}, English Fertility Score: {en_fertility_score} , Hindi Fertility Score: {hi_fertility_score}, Telugu Fertility Score: {ta_fertility_score}")

dataset_hi= load_dataset('ai4bharat/samanantar', 'hi', split='train', cache_dir='/sml1/atul/CENTRAL_CACHE')
dataset_ta= load_dataset('ai4bharat/samanantar', 'te', split='train', cache_dir='/sml1/atul/CENTRAL_CACHE')

test_en = dataset_hi['src'][:10000]
test_hi = dataset_hi['tgt'][:10000]
test_ta = dataset_ta['tgt'][:10000]

en_fertility_score=0
hi_fertility_score=0
ta_fertility_score=0

for data in test_en:
    tok=tokenizer(data)
    en_fertility_score += len(tok['input_ids']) / len(data.split())
en_fertility_score=en_fertility_score/10000

for data in test_hi:
    # print(data)
    tok=tokenizer(data)
    # print(tok)
    # exit()
    hi_fertility_score += len(tok['input_ids']) / len(data.split())
hi_fertility_score=hi_fertility_score/10000

for data in test_ta:
    tok=tokenizer(data)
    ta_fertility_score += len(tok['input_ids']) / len(data.split())
ta_fertility_score=ta_fertility_score/10000
log_parameters(64000, "Telugu-Llama7B", en_fertility_score, hi_fertility_score , ta_fertility_score )