[!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-indic-indic-1B-ONNX

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

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

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

ONNX-exported and quantized version of ai4bharat/indictrans2-indic-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	7.03 GB	100.00%	100.00%	100.00	94.7 ms	1.000x
FP16	3.52 GB	99.82%	99.82%	100.00	108.3 ms	0.874x
INT8	1.76 GB	83.64%	83.73%	94.22	43.7 ms	2.240x
Q4F16	900.0 MB	73.18%	73.18%	89.33	94.2 ms	1.087x

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	86.0%	86.0%	94.87	97.69
ben_Beng	50	96.0%	96.0%	98.16	99.26
brx_Deva	50	80.0%	80.0%	89.49	96.04
doi_Deva	50	96.0%	96.0%	99.04	99.07
gom_Deva	50	84.0%	84.0%	92.34	96.30
guj_Gujr	50	86.0%	86.0%	93.82	97.02
hin_Deva	50	88.0%	88.0%	94.85	96.34
kan_Knda	50	88.0%	88.0%	94.76	97.98
kas_Arab	50	88.0%	88.0%	94.88	96.68
mai_Deva	50	90.0%	90.0%	93.74	97.16
mal_Mlym	50	84.0%	86.0%	93.90	97.31
mar_Deva	50	86.0%	86.0%	94.54	97.21
mni_Beng	50	38.0%	38.0%	54.09	72.29
npi_Deva	50	78.0%	78.0%	87.37	93.85
ory_Orya	50	88.0%	88.0%	94.91	97.79
pan_Guru	50	94.0%	94.0%	96.84	98.00
san_Deva	50	84.0%	84.0%	90.62	96.80
sat_Olck	50	80.0%	80.0%	78.79	90.34
snd_Arab	50	66.0%	66.0%	82.16	87.27
tam_Taml	50	80.0%	80.0%	91.63	97.02
tel_Telu	50	92.0%	92.0%	95.85	97.76
urd_Arab	50	88.0%	88.0%	96.35	97.71

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	84.6%	84.6%	91.26	94.60
Lexicon	264	79.9%	80.3%	85.80	92.89
Numerals	264	84.1%	84.1%	93.63	96.73
Politics	286	85.7%	85.7%	92.32	96.38

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: Politics)

Source (asm_Beng → ben_Beng): অনুগ্ৰহ কৰি পৰৱৰ্তী সোমবাৰৰ ভিতৰত সম্পূৰ্ণ কৰা কামটো দাখিল কৰক।
Expected (FP32): দয়া করে পরবর্তী সোমবারের মধ্যে কাজটি সম্পন্ন করুন।
Actual (INT8): দয়া করে পরবর্তী সোমবারের মধ্যে কাজটি সম্পূর্ণ করে জমা দিন।

Mismatch #2 (Category: Lexicon)

Source (asm_Beng → ben_Beng): কোম্পানীটোৱে এক বন্ধুত্বপূৰ্ণ আৰু সহায়ক কৰ্ম পৰিৱেশ প্ৰদান কৰে।
Expected (FP32): সংস্থাটি একটি বন্ধুত্বপূর্ণ এবং সহায়ক কাজের পরিবেশ প্রদান করে।
Actual (INT8): কোম্পানিটি একটি বন্ধুত্বপূর্ণ এবং সহায়ক কাজের পরিবেশ প্রদান করে।

Mismatch #3 (Category: Lexicon)

Source (ben_Beng → brx_Deva): পেট্রোলের দাম পাঁচ টাকা বেড়েছে।
Expected (FP32): पेट्रलनि बेसेना से रां बांदों।
Actual (INT8): पेट्रलनि बेसेना रू।

Mismatch #4 (Category: Generic)

Source (ben_Beng → brx_Deva): ভালো বৃষ্টিপাতের কারণে ফসলের ফলন বেড়েছে।
Expected (FP32): मोजां अखानि जाहोनाव आबाद दिहुनथाया बांदों।
Actual (INT8): मोजां अखाया आबाद दिहुनथायाव बांहोदोंमोन।

Mismatch #5 (Category: Generic)

Source (ben_Beng → brx_Deva): আগামী মাস থেকে এই নতুন নীতি কার্যকর হবে।
Expected (FP32): गोदान खान्थिया फैगौ दाननिफ्राय बाहायजागासिनो।
Actual (INT8): गोदान खान्थिया फैगौ दाननिफ्राय बाहायजागोन।

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-indic-indic-1B-ONNX-int8")
print(model.translate("चुनाव कौन जीतेगा?", src_lang="hin_Deva", tgt_lang="tam_Taml"))

Required packages:

pip install onnxruntime tokenizers huggingface-hub

License

MIT (preserved from upstream AI4Bharat).