OpenVINO Tokenizers

OpenVINO Tokenizers adds text processing operations to OpenVINO.

Features

Perform tokenization and detokenization without third-party dependencies
Convert a HuggingFace tokenizer into OpenVINO model tokenizer and detokenizer
Combine OpenVINO models into a single model
Add greedy decoding pipeline to text generation model

Installation

(Recommended) Create and activate virtual env:

python3 -m venv venv
source venv/bin/activate
 # or
conda create --name openvino_tokenizers
conda activate openvino_tokenizers

Minimal Installation

Use minimal installation when you have a converted OpenVINO tokenizer:

pip install openvino-tokenizers
 # or
conda install -c conda-forge openvino openvino-tokenizers

Convert Tokenizers Installation

If you want to convert HuggingFace tokenizers into OpenVINO tokenizers:

pip install openvino-tokenizers[transformers]
 # or
conda install -c conda-forge openvino openvino-tokenizers && pip install transformers[sentencepiece] tiktoken

Install Pre-release Version

Use openvino-tokenizers[transformers] to install tokenizers conversion dependencies.

pip install --pre -U openvino openvino-tokenizers --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

Build and Install from Source

Using OpenVINO PyPI package

openvino-tokenizers build depends on openvino package which will be automatically installed from PyPI during the build process. To install unreleased versions, you would need to install openvino package from the nightly distribution channel using --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

git clone https://github.com/openvinotoolkit/openvino_tokenizers.git
cd openvino_tokenizers
pip install . --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

This command is the equivalent of minimal installation. Install tokenizers conversion dependencies if needed:

pip install transformers[sentencepiece] tiktoken

:warning: Latest commit of OpenVINO Tokenizers might rely on features that are not present in the release OpenVINO version. Use a nightly build of OpenVINO or build OpenVINO Tokenizers from a release branch if you have issues with the build process.

Using OpenVINO archive

Install OpenVINO archive distribution. Use --no-deps to avoid OpenVINO installation from PyPI into your current environment. --extra-index-url is needed to resolve build dependencies only.

source path/to/installed/openvino/setupvars.sh
git clone https://github.com/openvinotoolkit/openvino_tokenizers.git
cd openvino_tokenizers
pip install --no-deps . --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

This command is the equivalent of minimal installation. Install tokenizers conversion dependencies if needed:

pip install transformers[sentencepiece] tiktoken

Build and install for development

Using OpenVINO PyPI package

git clone https://github.com/openvinotoolkit/openvino_tokenizers.git
cd openvino_tokenizers
pip install -e .[all] --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly
# verify installation by running tests
cd tests/
pytest .

Using OpenVINO archive

Install OpenVINO archive distribution. Use --no-deps to avoid OpenVINO installation from PyPI into your current environment. --extra-index-url is needed to resolve build dependencies only.

source path/to/installed/openvino/setupvars.sh
git clone https://github.com/openvinotoolkit/openvino_tokenizers.git
cd openvino_tokenizers
pip install -e .[all] --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly
# verify installation by running tests
cd tests/
pytest .

C++ Installation

You can use converted tokenizers in C++ pipelines with prebuild binaries.

Download OpenVINO archive distribution for your OS from here and extract the archive.
Download OpenVINO Tokenizers prebuild libraries from here. To ensure compatibility first three numbers of OpenVINO Tokenizers version should match OpenVINO version and OS.
Extract OpenVINO Tokenizers archive into OpenVINO installation directory. OpenVINO Tokenizers archive maintains the structure to be aligned with OpenVINO archive:
- Windows: <openvino_dir>\runtime\bin\intel64\Release\
- MacOS_x86: <openvino_dir>/runtime/lib/intel64/Release
- MacOS_arm64: <openvino_dir>/runtime/lib/arm64/Release/
- Linux_x86: <openvino_dir>/runtime/lib/intel64/
- Linux_arm64: <openvino_dir>/runtime/lib/aarch64/

After that you can add binary extension in the code with:

core.add_extension("openvino_tokenizers.dll") for Windows
core.add_extension("libopenvino_tokenizers.dylib") for MacOS
core.add_extension("libopenvino_tokenizers.so") for Linux

and read/compile converted (de)tokenizers models. If you use version 2023.3.0.0, the binary extension file is called (lib)user_ov_extension.(dll/dylib/so).

C++ Build

To build OpenVINO Tokenizers binaries locally, use this command:

source path/to/installed/openvino/setupvars.sh
git clone https://github.com/openvinotoolkit/openvino_tokenizers.git
cd openvino_tokenizers
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make

After that, you can transfer all binaries from build/src to <openvino_dir> as described in the C++ installation instruction above.

Usage

:warning: OpenVINO Tokenizers can be inferred on a CPU device only.

Convert HuggingFace tokenizer

OpenVINO Tokenizers ships with CLI tool that can convert tokenizers from Huggingface Hub or Huggingface tokenizers saved on disk:

convert_tokenizer codellama/CodeLlama-7b-hf --with-detokenizer -o output_dir

There is also convert_tokenizer function that can convert tokenizer python object.

import numpy as np
from transformers import AutoTokenizer
from openvino import compile_model, save_model
from openvino_tokenizers import convert_tokenizer

hf_tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
ov_tokenizer = convert_tokenizer(hf_tokenizer)

compiled_tokenzier = compile_model(ov_tokenizer)
text_input = ["Test string"]

hf_output = hf_tokenizer(text_input, return_tensors="np")
ov_output = compiled_tokenzier(text_input)

for output_name in hf_output:
    print(f"OpenVINO {output_name} = {ov_output[output_name]}")
    print(f"HuggingFace {output_name} = {hf_output[output_name]}")
# OpenVINO input_ids = [[ 101 3231 5164  102]]
# HuggingFace input_ids = [[ 101 3231 5164  102]]
# OpenVINO token_type_ids = [[0 0 0 0]]
# HuggingFace token_type_ids = [[0 0 0 0]]
# OpenVINO attention_mask = [[1 1 1 1]]
# HuggingFace attention_mask = [[1 1 1 1]]

# save tokenizer for later use
save_model(ov_tokenizer, "openvino_tokenizer.xml")

loaded_tokenizer = compile_model("openvino_tokenizer.xml")
loaded_ov_output = loaded_tokenizer(text_input)
for output_name in hf_output:
    assert np.all(loaded_ov_output[output_name] == ov_output[output_name])

Connect Tokenizer to a Model

To infer and convert the original model, install torch or torch-cpu to the virtual environment.

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from openvino import compile_model, convert_model
from openvino_tokenizers import convert_tokenizer, connect_models

checkpoint = "mrm8488/bert-tiny-finetuned-sms-spam-detection"
hf_tokenizer = AutoTokenizer.from_pretrained(checkpoint)
hf_model = AutoModelForSequenceClassification.from_pretrained(checkpoint)

text_input = ["Free money!!!"]
hf_input = hf_tokenizer(text_input, return_tensors="pt")
hf_output = hf_model(**hf_input)

ov_tokenizer = convert_tokenizer(hf_tokenizer)
ov_model = convert_model(hf_model, example_input=hf_input.data)
combined_model = connect_models(ov_tokenizer, ov_model)
compiled_combined_model = compile_model(combined_model)

openvino_output = compiled_combined_model(text_input)

print(f"OpenVINO logits: {openvino_output['logits']}")
# OpenVINO logits: [[ 1.2007061 -1.4698029]]
print(f"HuggingFace logits {hf_output.logits}")
# HuggingFace logits tensor([[ 1.2007, -1.4698]], grad_fn=<AddmmBackward0>)

Use Extension With Converted (De)Tokenizer or Model With (De)Tokenizer

Import openvino_tokenizers will register tokenizer-related operations to OpenVINO, after which you can work with saved tokenizers and detokenizers.

import numpy as np
import openvino_tokenizers
from openvino import Core

core = Core()

# detokenizer from codellama sentencepiece model
compiled_detokenizer = core.compile_model("detokenizer.xml")

token_ids = np.random.randint(100, 1000, size=(3, 5))
openvino_output = compiled_detokenizer(token_ids)

print(openvino_output["string_output"])
# ['sc�ouition�', 'intvenord hasient', 'g shouldwer M more']

Text Generation Pipeline

import numpy as np
from openvino import compile_model, convert_model
from openvino_tokenizers import add_greedy_decoding, convert_tokenizer
from transformers import AutoModelForCausalLM, AutoTokenizer


model_checkpoint = "JackFram/llama-68m"
hf_tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
hf_model = AutoModelForCausalLM.from_pretrained(model_checkpoint, use_cache=False)

# convert hf tokenizer
text_input = ["Quick brown fox jumped "]
ov_tokenizer, ov_detokenizer = convert_tokenizer(hf_tokenizer, with_detokenizer=True)
compiled_tokenizer = compile_model(ov_tokenizer)

# transform input text into tokens
ov_input = compiled_tokenizer(text_input)
hf_input = hf_tokenizer(text_input, return_tensors="pt")

# convert Pytorch model to OpenVINO IR and add greedy decoding pipeline to it
ov_model = convert_model(hf_model, example_input=hf_input.data)
ov_model_with_greedy_decoding = add_greedy_decoding(ov_model)
compiled_model = compile_model(ov_model_with_greedy_decoding)

# generate new tokens
new_tokens_size = 10
prompt_size = ov_input["input_ids"].shape[-1]
input_dict = {
    output.any_name: np.hstack([tensor, np.zeros(shape=(1, new_tokens_size), dtype=np.int_)])
    for output, tensor in ov_input.items()
}
for idx in range(prompt_size, prompt_size + new_tokens_size):
    output = compiled_model(input_dict)["token_ids"]
    input_dict["input_ids"][:, idx] = output[:, idx - 1]
    input_dict["attention_mask"][:, idx] = 1
ov_token_ids = input_dict["input_ids"]

hf_token_ids = hf_model.generate(
    **hf_input,
    min_new_tokens=new_tokens_size,
    max_new_tokens=new_tokens_size,
    temperature=0,  # greedy decoding
)

# decode model output
compiled_detokenizer = compile_model(ov_detokenizer)
ov_output = compiled_detokenizer(ov_token_ids)["string_output"]
hf_output = hf_tokenizer.batch_decode(hf_token_ids, skip_special_tokens=True)
print(f"OpenVINO output string: `{ov_output}`")
# OpenVINO output string: `['Quick brown fox was walking through the forest. He was looking for something']`
print(f"HuggingFace output string: `{hf_output}`")
# HuggingFace output string: `['Quick brown fox was walking through the forest. He was looking for something']`

TensorFlow Text Integration

OpenVINO Tokenizers include converters for certain TensorFlow Text operations. Currently, only the MUSE model is supported. Here is an example of model conversion and inference:

import numpy as np
import tensorflow_hub as hub
import tensorflow_text  # register tf text ops
from openvino import convert_model, compile_model
import openvino_tokenizers  # register ov tokenizer ops and translators


sentences = ["dog",  "I cuccioli sono carini.", "私は犬と一緒にビーチを散歩するのが好きです"]
tf_embed = hub.load(
    "https://www.kaggle.com/models/google/universal-sentence-encoder/frameworks/"
    "TensorFlow2/variations/multilingual/versions/2"
)
# convert model that uses Sentencepiece tokenizer op from TF Text
ov_model = convert_model(tf_embed)
ov_embed = compile_model(ov_model, "CPU")

ov_result = ov_embed(sentences)[ov_embed.output()]
tf_result = tf_embed(sentences)

assert np.all(np.isclose(ov_result, tf_result, atol=1e-4))

RWKV Tokenizer

from urllib.request import urlopen

from openvino import compile_model
from openvino_tokenizers import build_rwkv_tokenizer


rwkv_vocab_url = (
    "https://raw.githubusercontent.com/BlinkDL/ChatRWKV/main/tokenizer/rwkv_vocab_v20230424.txt"
)

with urlopen(rwkv_vocab_url) as vocab_file:
    vocab = map(bytes.decode, vocab_file)
    tokenizer, detokenizer = build_rwkv_tokenizer(vocab)

tokenizer, detokenizer = compile_model(tokenizer), compile_model(detokenizer)

print(tokenized := tokenizer(["Test string"])["input_ids"])  # [[24235 47429]]
print(detokenizer(tokenized)["string_output"])  # ['Test string']

Tokenizer From GGUF Model

from transformers import AutoTokenizer
import openvino as ov
from openvino_tokenizers import convert_tokenizer


model_id = "unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF"
filename = "DeepSeek-R1-Distill-Qwen-1.5B-Q2_K.gguf"
hf_tokenizer = AutoTokenizer.from_pretrained(model_id, gguf_file=filename)

ov_tokenizer, ov_detokenizer = convert_tokenizer(hf_tokenizer, with_detokenizer=True)
ov_tokenizer, ov_detokenizer = ov.compile_model(ov_tokenizer), ov.compile_model(ov_detokenizer)

print(ov_res := ov_tokenizer(["Test string"])["input_ids"])  # [[2271  914]]
print(ov_detokenizer(ov_res)["string_output"])  # ['Test string']

C++ Usage Example

This example shows how to run inference with C++ on a text-classification model from Hugging Face. It expects the path to a model directory as parameter, and prints the logits returned by the model inference.

Export an example model by running the following command after pip install optimum[openvino]:

optimum-cli export openvino microsoft/deberta-base-mnli deberta-base-mnli-ov

#include <openvino/openvino.hpp>
#include <iostream>
#include <filesystem>

int main(int argc, char* argv[]) {
   std::string dirname = argv[1];
   std::filesystem::path dir_path(dirname);
   std::filesystem::path model_xml = dir_path / "openvino_model.xml";
   std::filesystem::path tokenizer_xml = dir_path / "openvino_tokenizer.xml";

   ov::Core core;
   // use "openvino_tokenizers.dll" on Windows, "libopenvino_tokenizers.dylib" on macOS
   core.add_extension("libopenvino_tokenizers.so");

   ov::InferRequest tokenizer_request = core.compile_model(tokenizer_xml, "CPU").create_infer_request();

   std::string prompt="Hello world!";
   tokenizer_request.set_input_tensor(ov::Tensor{ov::element::string, {1}, &prompt});
   tokenizer_request.infer();
   ov::Tensor input_ids = tokenizer_request.get_tensor("input_ids");
   ov::Tensor attention_mask = tokenizer_request.get_tensor("attention_mask");

   ov::InferRequest infer_request = core.compile_model(model_xml, "CPU").create_infer_request();
   infer_request.set_tensor("input_ids", input_ids);
   infer_request.set_tensor("attention_mask", attention_mask);
   infer_request.infer();

   auto output = infer_request.get_tensor("logits");
   const float *output_buffer = output.data<const float>();

   size_t num_elements = output.get_size();

   for (size_t i = 0; i < num_elements; i++) {
       std::cout << output_buffer[i] << " ";
   }

   std::cout << std::endl;
   return 0;
}

Unicode Support

OpenVINO Tokenizers support UTF-8 encoded inputs.
Internal tokenizer vocabulary is stored in UTF-8 encoding:
- Providing a tokenizer model with non-UTF-8 input may lead to unexpected outputs or errors,
- Detokenizer output is UTF-8 encoded; if your terminal does not expect UTF-8, you might see garbage characters.
By default, a detokenizer replaces invalid UTF-8 output with � character. You can change this behavior during conversion.

Supported Tokenizer Types

Huggingface Tokenizer Type	Tokenizer Model Type	Tokenizer	Detokenizer
Fast	WordPiece	✅	✅
	BPE	✅	✅
	Unigram	✅	✅
	WordLevel*	✅	✅
Legacy	SentencePiece .model	✅	✅
Custom	tiktoken	✅	✅
RWKV	Trie	✅	✅

Test Results

This report is autogenerated and includes tokenizers and detokenizers tests. The Output Matched, % column shows the percent of test strings for which the results of OpenVINO and Huggingface Tokenizers are the same. To update the report run pytest --update_readme tokenizers_test.py in tests directory.

Output Match by Tokenizer Type

Tokenizer Type	Output Matched, %	Number of Tests
BPE	99.26	6218
SentencePiece	89.50	6036
Tiktoken	96.64	536
Unigram	95.35	1506
WordLevel	98.99	198
WordPiece	99.09	1319

Output Match by Model

Tokenizer Type	Model	Output Matched, %	Number of Tests
BPE	NousResearch/Llama-2-13b-hf	97.61	251
BPE	NousResearch/Meta-Llama-3-8B-Instruct	100.00	253
BPE	Qwen/Qwen3-Reranker-0.6B	100.00	269
BPE	Salesforce/codegen-16B-multi	100.00	267
BPE	TinyLlama/TinyLlama-1.1B-Chat-v1.0	100.00	253
BPE	Xenova/gpt-4o	100.00	267
BPE	ai-forever/rugpt3large_based_on_gpt2	100.00	267
BPE	allenai/OLMo-1B-hf	100.00	251
BPE	answerdotai/ModernBERT-base	100.00	267
BPE	bigscience/bloom	97.61	251
BPE	deepseek-ai/DeepSeek-V3-0324	99.26	269
BPE	deepseek-ai/deepseek-coder-6.7b-instruct	100.00	269
BPE	facebook/galactica-120b	100.00	251
BPE	koalajun/Gemma-2-9b-it-Ko-Crypto-Translate	100.00	253
BPE	laion/CLIP-ViT-bigG-14-laion2B-39B-b160k	100.00	267
BPE	llava-hf/LLaVA-NeXT-Video-7B-hf	97.61	251
BPE	microsoft/Phi-3-mini-128k-instruct	100.00	253
BPE	microsoft/deberta-base	100.00	251
BPE	mlx-community/quantized-gemma-7b-it	97.63	253
BPE	roberta-base	100.00	267
BPE	stabilityai/stablecode-completion-alpha-3b-4k	100.00	251
BPE	stabilityai/stablelm-2-1_6b	100.00	251
BPE	tiiuae/Falcon3-7B-Instruct	96.28	269
BPE	tiiuae/falcon-7b	96.25	267
SentencePiece	BAAI/bge-reranker-v2-m3	96.81	251
SentencePiece	BAAI/bge-reranker-v2-m3_legacy	96.81	251
SentencePiece	NousResearch/Llama-2-13b-hf	94.42	251
SentencePiece	NousResearch/Llama-2-13b-hf_legacy	97.61	251
SentencePiece	TinyLlama/TinyLlama-1.1B-Chat-v1.0	100.00	253
SentencePiece	TinyLlama/TinyLlama-1.1B-Chat-v1.0_legacy	98.42	253
SentencePiece	baichuan-inc/Baichuan2-7B-Chat_legacy	100.00	251
SentencePiece	camembert-base	56.18	251
SentencePiece	camembert-base_legacy	78.88	251
SentencePiece	facebook/musicgen-small	82.07	251
SentencePiece	facebook/musicgen-small_legacy	76.10	251
SentencePiece	google/flan-t5-xxl	75.70	251
SentencePiece	google/flan-t5-xxl_legacy	74.50	251
SentencePiece	llava-hf/LLaVA-NeXT-Video-7B-hf	93.63	251
SentencePiece	llava-hf/LLaVA-NeXT-Video-7B-hf_legacy	96.81	251
SentencePiece	microsoft/Phi-3-mini-128k-instruct	99.21	253
SentencePiece	microsoft/Phi-3-mini-128k-instruct_legacy	97.63	253
SentencePiece	microsoft/deberta-v3-base	95.22	251
SentencePiece	microsoft/deberta-v3-base_legacy	98.41	251
SentencePiece	microsoft/speecht5_tts_legacy	71.71	251
SentencePiece	mlx-community/quantized-gemma-7b-it	96.84	253
SentencePiece	mlx-community/quantized-gemma-7b-it_legacy	97.63	253
SentencePiece	rinna/bilingual-gpt-neox-4b	83.27	251
SentencePiece	rinna/bilingual-gpt-neox-4b_legacy	89.64	251
Tiktoken	Qwen/Qwen-14B-Chat	100.00	267
Tiktoken	THUDM/glm-4-9b-chat	93.31	269
Unigram	BAAI/bge-reranker-v2-m3	98.41	251
Unigram	camembert-base	84.86	251
Unigram	facebook/musicgen-small	98.41	251
Unigram	google/flan-t5-xxl	92.03	251
Unigram	microsoft/deberta-v3-base	98.41	251
Unigram	rinna/bilingual-gpt-neox-4b	100.00	251
WordLevel	cisco-ai/mini-bart-g2p	98.99	198
WordPiece	bert-base-multilingual-cased	100.00	267
WordPiece	cointegrated/rubert-tiny2	100.00	267
WordPiece	google/mobilebert-uncased	100.00	251
WordPiece	rasa/LaBSE	95.51	267
WordPiece	sentence-transformers/all-MiniLM-L6-v2	100.00	267

Recreating Tokenizers From Tests

In some tokenizers, you need to select certain settings so that their output is closer to the Huggingface tokenizers:

THUDM/chatglm3-6b detokenizer don't skips special tokens. Use skip_special_tokens=False during conversion
All tested tiktoken based detokenizers leave extra spaces. Use clean_up_tokenization_spaces=False during conversion

Oven