language:
- as
- bn
- brx
- doi
- gom
- gu
- hi
- kn
- ks
- mai
- ml
- mr
- mni
- ne
- or
- pa
- sa
- sat
- snd
- ta
- te
- ur
language_details: >-
asm_Beng, ben_Beng, brx_Deva, doi_Deva, gom_Deva, guj_Gujr,
hin_Deva, kan_Knda, kas_Arab, mai_Deva, mal_Mlym, mar_Deva,
mni_Mtei, npi_Deva, ory_Orya, pan_Guru, san_Deva, sat_Olck,
snd_Deva, tam_Taml, tel_Telu, urd_Arab
tags:
- indictrans2
- translation
- ai4bharat
- multilingual
license: mit
datasets:
- flores-200
- IN22-Gen
- IN22-Conv
metrics:
- bleu
- chrf
- chrf++
- comet
inference: false
IndicTrans2
This is the model card of IndicTrans2 Indic-Indic Distilled 320M variant adapted after stitching Indic-En Distilled 200M and En-Indic Distilled 200M variants.
Please refer to the blog for further details on model training, data and metrics.
Usage Instructions
Please refer to the github repository for a detail description on how to use HF compatible IndicTrans2 models for inference.
import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
from IndicTransToolkit.processor import IndicProcessor
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
src_lang, tgt_lang = "hin_Deva", "tam_Taml"
model_name = "ai4bharat/indictrans2-indic-indic-dist-320M"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForSeq2SeqLM.from_pretrained(
model_name,
trust_remote_code=True,
torch_dtype=torch.float16,
attn_implementation="flash_attention_2"
).to(DEVICE)
ip = IndicProcessor(inference=True)
input_sentences = [
"जब मैं छोटा था, मैं हर रोज़ पार्क जाता था।",
"हमने पिछले सप्ताह एक नई फिल्म देखी जो कि बहुत प्रेरणादायक थी।",
"अगर तुम मुझे उस समय पास मिलते, तो हम बाहर खाना खाने चलते।",
"मेरे मित्र ने मुझे उसके जन्मदिन की पार्टी में बुलाया है, और मैं उसे एक तोहफा दूंगा।",
]
batch = ip.preprocess_batch(
input_sentences,
src_lang=src_lang,
tgt_lang=tgt_lang,
)
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
inputs = tokenizer(
batch,
truncation=True,
padding="longest",
return_tensors="pt",
return_attention_mask=True,
).to(DEVICE)
with torch.no_grad():
generated_tokens = model.generate(
**inputs,
use_cache=True,
min_length=0,
max_length=256,
num_beams=5,
num_return_sequences=1,
)
generated_tokens = tokenizer.batch_decode(
generated_tokens,
skip_special_tokens=True,
clean_up_tokenization_spaces=True,
)
translations = ip.postprocess_batch(generated_tokens, lang=tgt_lang)
for input_sentence, translation in zip(input_sentences, translations):
print(f"{src_lang}: {input_sentence}")
print(f"{tgt_lang}: {translation}")
Citation
If you consider using our work then please cite using:
@article{gala2023indictrans,
title={IndicTrans2: Towards High-Quality and Accessible Machine Translation Models for all 22 Scheduled Indian Languages},
author={Jay Gala and Pranjal A Chitale and A K Raghavan and Varun Gumma and Sumanth Doddapaneni and Aswanth Kumar M and Janki Atul Nawale and Anupama Sujatha and Ratish Puduppully and Vivek Raghavan and Pratyush Kumar and Mitesh M Khapra and Raj Dabre and Anoop Kunchukuttan},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2023},
url={https://openreview.net/forum?id=vfT4YuzAYA},
note={}
}