Oven logo

Oven

OpenVINO Tokenizers

Downloads Anaconda-Server Badge

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

: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.

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.

  1. Download OpenVINO archive distribution for your OS from here and extract the archive.
  2. Download OpenVINO Tokenizers prebuild libraries from here. To ensure compatibility first three numbers of OpenVINO Tokenizers version should match OpenVINO version and OS.
  3. 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).

Reducing the ICU Data Size

By default, all available ICU locales are supported, which significantly increases the package size. To reduce the size of the ICU libraries included in your final package, follow these steps:

  1. Use the ICU Data Configuration File:

    • This file specifies which features and locales to include in a custom data bundle. You can find more information here.
  2. Set the ICU Data Filter File as an Environment Variable:

    • On Unix-like systems (Linux, macOS): Set the ICU_DATA_FILTER_FILE environment variable to the path of your configuration file (filters.json):

      export ICU_DATA_FILTER_FILE="filters.json"
      
    • On Windows: Set the ICU_DATA_FILTER_FILE environment variable using the Command Prompt or PowerShell:

      Command Prompt:

      set ICU_DATA_FILTER_FILE=filters.json
      

      PowerShell:

      $env:ICU_DATA_FILTER_FILE="filters.json"
      
  3. Create a Configuration File:

    • An example configuration file (filters.json) might look like this:
    {
      "localeFilter": {
        "filterType": "language",
        "includelist": [
          "en"
        ]
      }
    }
    
  4. Configure OpenVINO Tokenizers:

    • When building OpenVINO tokenizers, set the following CMake option during the project configuration:
    -DBUILD_FAST_TOKENIZERS=ON
    
    • Example for a pip installation path:
    ICU_DATA_FILTER_FILE=</path/to/filters.json> pip install git+https://github.com/openvinotoolkit/openvino_tokenizers.git --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly --config-settings=override=cmake.options.BUILD_FAST_TOKENIZERS=ON
    

By following these instructions, you can effectively reduce the size of the ICU libraries in your final package.

Build OpenVINO Tokenizers without FastTokenizer Library

If a tokenizer doesn't use CaseFold, UnicodeNormalization or Wordpiece operations, you can drastically reduce package binary size by building OpenVINO Tokenizers without FastTokenizer dependency with this flag:

-DENABLE_FAST_TOKENIZERS=OFF

This option can also help with building for platform that is supported by FastTokenizer, for example Android x86_64.

Example for a pip installation path:


pip install git+https://github.com/openvinotoolkit/openvino_tokenizers.git --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly --config-settings=override=cmake.options.ENABLE_FAST_TOKENIZERS=OFF

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 add all 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']

Supported Tokenizer Types

Huggingface
Tokenizer Type
Tokenizer Model TypeTokenizerDetokenizer
FastWordPiece
BPE
Unigram
LegacySentencePiece .model
Customtiktoken
RWKVTrie

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 TypeOutput Matched, %Number of Tests
BPE95.575932
SentencePiece88.236534
Tiktoken99.19494
WordPiece99.101327

Output Match by Model

Tokenizer TypeModelOutput Matched, %Number of Tests
BPEEleutherAI/gpt-j-6b95.29255
BPEEleutherAI/gpt-neo-125m95.29255
BPEEleutherAI/gpt-neox-20b95.82239
BPEEleutherAI/pythia-12b-deduped95.82239
BPEKoboldAI/fairseq-dense-13B96.65239
BPENousResearch/Meta-Llama-3-8B-Instruct100.00241
BPESalesforce/codegen-16B-multi96.08255
BPEXenova/gpt-4o100.00255
BPEai-forever/rugpt3large_based_on_gpt294.51255
BPEbigscience/bloom97.49239
BPEdatabricks/dolly-v2-3b95.82239
BPEdeepseek-ai/deepseek-coder-6.7b-instruct100.00257
BPEfacebook/bart-large-mnli95.29255
BPEfacebook/galactica-120b95.82239
BPEfacebook/opt-66b96.65239
BPEgpt295.29255
BPElaion/CLIP-ViT-bigG-14-laion2B-39B-b160k75.29255
BPEmicrosoft/deberta-base96.65239
BPEroberta-base95.29255
BPEsentence-transformers/all-roberta-large-v195.29255
BPEstabilityai/stablecode-completion-alpha-3b-4k95.82239
BPEstabilityai/stablelm-2-1_6b100.00239
BPEstabilityai/stablelm-tuned-alpha-7b95.82239
BPEtiiuae/falcon-7b94.51255
SentencePieceNousResearch/Llama-2-13b-hf96.65239
SentencePieceNousResearch/Llama-2-13b-hf_legacy100.00239
SentencePieceNousResearch/Llama-2-13b-hf_sp_backend100.00239
SentencePieceTHUDM/chatglm2-6b_legacy100.00153
SentencePieceTHUDM/chatglm3-6b_legacy50.97155
SentencePiececamembert-base52.30239
SentencePiececamembert-base_legacy76.15239
SentencePiececodellama/CodeLlama-7b-hf96.65239
SentencePiececodellama/CodeLlama-7b-hf_legacy96.65239
SentencePiececodellama/CodeLlama-7b-hf_sp_backend94.98239
SentencePiecefacebook/musicgen-small84.52239
SentencePiecefacebook/musicgen-small_legacy79.92239
SentencePiecemicrosoft/Phi-3-mini-128k-instruct95.85241
SentencePiecemicrosoft/Phi-3-mini-128k-instruct_legacy95.85241
SentencePiecemicrosoft/Phi-3-mini-128k-instruct_sp_backend94.19241
SentencePiecemicrosoft/deberta-v3-base96.65239
SentencePiecemicrosoft/deberta-v3-base_legacy100.00239
SentencePiecemlx-community/quantized-gemma-7b-it99.17241
SentencePiecemlx-community/quantized-gemma-7b-it_legacy99.17241
SentencePiecemlx-community/quantized-gemma-7b-it_sp_backend100.00241
SentencePiecerinna/bilingual-gpt-neox-4b80.75239
SentencePiecerinna/bilingual-gpt-neox-4b_legacy86.61239
SentencePiecet5-base85.77239
SentencePiecet5-base_legacy81.17239
SentencePiecexlm-roberta-base96.23239
SentencePiecexlm-roberta-base_legacy96.23239
SentencePiecexlnet-base-cased65.27239
SentencePiecexlnet-base-cased_legacy59.41239
TiktokenQwen/Qwen-14B-Chat100.00255
TiktokenTHUDM/glm-4-9b98.33239
WordPieceProsusAI/finbert100.00107
WordPiecebert-base-multilingual-cased100.00107
WordPiecebert-base-uncased100.00107
WordPiececointegrated/rubert-tiny2100.00107
WordPiecedistilbert-base-uncased-finetuned-sst-2-english100.00107
WordPiecegoogle/electra-base-discriminator100.00107
WordPiecegoogle/mobilebert-uncased100.0091
WordPiecejhgan/ko-sbert-sts100.00107
WordPieceprajjwal1/bert-mini100.0091
WordPiecerajiv003/ernie-finetuned-qqp100.0091
WordPiecerasa/LaBSE88.79107
WordPiecesentence-transformers/all-MiniLM-L6-v2100.00107
WordPiecesqueezebert/squeezebert-uncased100.0091

Recreating Tokenizers From Tests

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

  • THUDM/chatglm2-6b detokenizer always skips special tokens. Use skip_special_tokens=True during conversion
  • 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