[!TIP] This model is part of a suite of optimized/quantized ONNX versions of the base model. Other variants in this direction:

FP32 (Full Precision / Base): hari31416/indictrans2-en-indic-1B-ONNX

FP16 (Half Precision): hari31416/indictrans2-en-indic-1B-ONNX-fp16

INT8 (Dynamic Quantization): hari31416/indictrans2-en-indic-1B-ONNX-int8 (Current)

Q4F16 (4-bit Block Quantization): hari31416/indictrans2-en-indic-1B-ONNX-q4f16

ONNX-exported and quantized version of ai4bharat/indictrans2-en-indic-1B for in-browser and local edge inference.

Precision: INT8 (Dynamic Quantization)
Description: Dynamic INT8 quantization of the encoder and decoder. Highly recommended for CPU environments.
Source Pipeline & Details: For pipeline details, benchmarks, and usage instructions, see the indictrans2-onnx-export GitHub repository.

Built for use with Transformers.js and onnxruntime-web in the browser, with fast BPE tokenizer.json files that don't require the SentencePiece WASM runtime.

Performance Visualizations

These charts show overall tradeoffs, language-level parity, and category breakdown.

Performance Tradeoffs & Size Comparison

Compared against the FP32 ONNX oracle on the golden evaluation fixtures.

Format	Model Size	Exact Match (Token)	Exact Match (Text)	SacreBLEU (Raw)	Latency (Mean)	Speedup vs. FP32
FP32	6.64 GB	100.00%	100.00%	100.00	69.5 ms	1.000x
FP16	3.32 GB	99.73%	99.73%	100.00	74.3 ms	0.935x
INT8	1.66 GB	89.55%	89.55%	96.27	31.4 ms	2.125x
Q4F16	850.5 MB	82.45%	82.55%	91.99	58.4 ms	1.186x

Language-Level Parity (INT8)

Exact match rates and translation quality (SacreBLEU / chrF) per language pair under this precision:

Language Code	Total Fixtures	Token Match Rate	Text Match Rate	SacreBLEU	SacreBLEU (chrF)
asm_Beng	50	92.0%	92.0%	97.28	98.99
ben_Beng	50	92.0%	92.0%	94.53	97.99
brx_Deva	50	80.0%	80.0%	91.68	97.20
doi_Deva	50	86.0%	86.0%	95.99	97.67
gom_Deva	50	94.0%	94.0%	96.83	99.47
guj_Gujr	50	90.0%	90.0%	96.73	98.74
hin_Deva	50	98.0%	98.0%	98.94	99.63
kan_Knda	50	88.0%	88.0%	94.90	98.17
kas_Arab	50	92.0%	92.0%	95.75	97.49
mai_Deva	50	92.0%	92.0%	95.54	98.14
mal_Mlym	50	100.0%	100.0%	100.00	100.00
mar_Deva	50	92.0%	92.0%	96.04	98.67
mni_Beng	50	72.0%	72.0%	84.69	92.04
npi_Deva	50	96.0%	96.0%	97.29	98.64
ory_Orya	50	90.0%	90.0%	94.87	98.94
pan_Guru	50	92.0%	92.0%	96.78	98.32
san_Deva	50	76.0%	76.0%	86.50	96.14
sat_Olck	50	66.0%	66.0%	83.16	90.39
snd_Arab	50	96.0%	96.0%	97.47	98.96
tam_Taml	50	94.0%	94.0%	96.50	98.95
tel_Telu	50	94.0%	94.0%	97.87	98.64
urd_Arab	50	98.0%	98.0%	99.36	99.69

Category-Level Parity (INT8)

Exact match rates and translation quality grouped by category types:

Category	Total Fixtures	Token Match Rate	Text Match Rate	SacreBLEU	SacreBLEU (chrF)
Generic	286	87.1%	87.1%	93.24	96.89
Lexicon	264	89.4%	89.4%	95.02	97.75
Numerals	264	90.5%	90.5%	95.92	98.21
Politics	286	91.3%	91.3%	95.94	98.11

Translation Mismatch Examples

Here is a sample of up to 5 translation mismatches compared to the FP32 oracle. Many mismatches represent minor synonym differences or spacing variations.

Mismatch #1 (Category: Numerals)

Source (eng_Latn → asm_Beng): She sang a beautiful song at the concert.
Expected (FP32): কনচাৰ্টটোত তেওঁ এটা সুন্দৰ গীত গাইছিল।
Actual (INT8): কনচাৰ্টখনত তেওঁ এটা সুন্দৰ গীত গাইছিল।

Mismatch #2 (Category: Lexicon)

Source (eng_Latn → asm_Beng): The books are arranged on the shelf in alphabetical order.
Expected (FP32): কিতাপবোৰ বৰ্ণানুক্রমিক ক্ৰমত তাকত সজোৱা হয়।
Actual (INT8): কিতাপবোৰ বৰ্ণানুক্রমিকভাৱে তাকত সজোৱা হয়।

Mismatch #3 (Category: Generic)

Source (eng_Latn → asm_Beng): A warm cup of tea is perfect for a cold morning.
Expected (FP32): এক গৰম কাপ চাহ ঠাণ্ডা ৰাতিপুৱাৰ বাবে উপযুক্ত।
Actual (INT8): ঠাণ্ডা ৰাতিপুৱাৰ বাবে এক গৰম কাপ চাহ উপযুক্ত।

Mismatch #4 (Category: Generic)

Source (eng_Latn → asm_Beng): The price of gold has reached a new high.
Expected (FP32): সোণৰ মূল্য নতুন উচ্চতাত উপনীত হৈছে।
Actual (INT8): সোণৰ মূল্য এক নতুন উচ্চতাত উপনীত হৈছে।

Mismatch #5 (Category: Generic)

Source (eng_Latn → ben_Beng): The quick brown fox jumps over the lazy dog.
Expected (FP32): দ্রুত বাদামী শিয়াল অলস কুকুরের উপর ঝাঁপিয়ে পড়ে।
Actual (INT8): দ্রুত বাদামী শিয়ালটি অলস কুকুরটির উপর ঝাঁপিয়ে পড়ে।

Files

encoder_model.onnx (and optional .onnx.data weights sidecar)
decoder_model.onnx and decoder_with_past_model.onnx (share decoder_shared.onnx.data when present)
translate.py — self-contained Python inference helper (see Usage below)
Fast tokenizer config files (tokenizer_src.json, tokenizer_tgt.json, tokenizer_meta.json)
Model configuration configs (config.json, generation_config.json)

Usage Example (Python, onnxruntime)

python

# translate.py is included in this repo alongside the ONNX bundle.
# You can also find it (and read the full source) at:
#   https://github.com/Hari31416/indictrans2-onnx-export/blob/main/src/translate.py

from translate import IndicTransONNX

# Pass a HF repo ID for automatic download, or a local bundle directory path
model = IndicTransONNX("hari31416/indictrans2-en-indic-1B-ONNX-int8")
print(model.translate("Who will win the election?", src_lang="eng_Latn", tgt_lang="hin_Deva"))

Required packages:

pip install onnxruntime tokenizers huggingface-hub

License

MIT (preserved from upstream AI4Bharat).