Skip to main content

LLMtranslator translates and generates text in multiple languages.

Project description

Development Status :: 3 - Alpha

Python Package: hf-transllm

PyPI package Code style: black PyPI

LLMtranslator translates and generates text in multiple languages using LLMs(Large Language Models) on hugging-face repository.

Introducing hf-transllm: Unlock the Power of Multilingual Exploration

Discover the hf-transllm package, a seamless integration of Hugging Face's inference module and translation APIs. Overcome limitations in retraining and evaluating large language models in different languages. Explore diverse results effortlessly, leveraging translation API services. Emphasizing versatility over efficiency, hf-transllm enables you to delve into the outcomes of Hugging Face's models in various languages.



Installation

pip install transllm

Or

pip install git+https://github.com/dsdanielpark/hf-trnasllm.git

CLI

If you wish to use CLI:

git clone https://github.com/dsdanielpark/hf-transllm
cd hf-transllm
pip install -r requirements.txt
python main.py --hfmodel <openlm-research/open_llama_3b> --lang <ko> --translator <google>

There can be issues with various dependencies such as Hugging Face's Transformers, SentencePiece, Torch, and CUDA. Please set up the appropriate environment by searching online.


Usage

Simple Usage

from transllm import LLMtranslator

open_llama3b_kor = LLMtranslator('openlm-research/open_llama_3b', target_lang='ko', translator='google') # Korean

trnaslated_answer = open_llama3b_kor.generate("나와 내 동년배들이 좋아하는 뉴진스에 대해서 알려줘")
print(trnaslated_answer)

Official Google Translation API

Unofficial Google Trnaslator for non-profit purposes (such as feature testing)

from transllm import LLMtranslator

# Set huggingface repository
model_path = 'openlm-research/open_llama_3b'
# model_path = 'openlm-research/open_llama_7b'
# model_path = 'openlm-research/open_llama_13b'

# Get TransLLM Object (Korean)
# open_llama3b_kor = LLMtranslator(model_path, target_lang='ko', translator='google_official', google_api_key='xxxxxx') # Official Google Cloud Translation API 
open_llama3b_kor = LLMtranslator(model_path, target_lang='ko', translator='google') # Unofficial test



# Using Prompt in multi-language
prompt = "나와 내 동년배들이 좋아하는 뉴진스에 대해서 알려줘"
trnaslated_answer = open_llama3b_kor.generate(prompt)
print(trnaslated_answer)

DeepL, Open AI, Bard

Open AI, Bard use pre-prompt for translation.

from transllm import LLMtranslator

# Set huggingface repository
model_path = 'openlm-research/open_llama_3b'
# model_path = 'openlm-research/open_llama_7b'
# model_path = 'openlm-research/open_llama_13b'

# Choose Translate Service API: DeepL, OpenAI, Bard
open_llama3b_kor = LLMtranslator(model_path, target_lang='ES', translator='deepl', deepl_api='xxxxxxx') # Language == Spanish
# open_llama3b_kor = LLMtranslator(model_path, target_lang='korean', translator='openai', openai_api='xxxxxxx', openai_model='gpt-3.5-turbo')
# open_llama3b_kor = LLMtranslator(model_path, target_lang='korean', translator='bard', bard_api='xxxxxxx')

# Using Prompt in multi-language
prompt = "나와 내 동년배들이 좋아하는 뉴진스에 대해서 알려줘"
trnaslated_answer = open_llama3b_kor.generate(prompt)
print(trnaslated_answer)

Google Colab Example

Open In Colab



Customized Inference

Applying LLMs to the majority of Hugging Face repositories is generally feasible. However, it can be challenging to apply them to objects that require unique tokenizers or inference processes. In such cases, it is recommended to customize the usage by incorporating a translation module for prompts.

In other words, if you are familiar with the inference process or code from Hugging Face repositories, you can customize the translation object by adding a translation module before and after the known inference process or code. Check translang repository for more details.

import torch
from trnasllm import LLMtranslator


class MyLLMtranslator(LLMtranslator):
    def __init__(
        self,
        model_path,
        target_lang="ko",
        translator="google",
        torch_dtype=torch.float16,
        device_map="auto",
        deepl_api=None,
        bard_api=None,
        openai_model="gpt-3.5-turbo",
        openai_api=None
    ):
        super().__init__(
            model_path=model_path,
            target_lang=target_lang,
            translator=translator,
            torch_dtype=torch_dtype,
            device_map=device_map,
            deepl_api=deepl_api,
            bard_api=bard_api,
            openai_model=openai_model,
            openai_api=openai_api
        )

    def inference(self, prompt: str) -> str:
        # Custom logic for inference
        # You can override the implementation of the inference method here
        # and provide your own logic for generating the translated answer
        # Remember to return the translated answer as a string.

        answer = customizing_process(prompt)
        # Custom inference logic...
        
        return answer

About Google Translator

Commercial use or official use of the Google Translate service is chargeable. Please provide the translator="google_official" and google_api_key={YOUR_API_KEY} arguments. Please responsibly use the translator="google" argument for the purpose of simple functionality verification. Refer to the following notebook file and official link for more information. Use the google argument only for some basic functionality testing.

FAQs

Use Ctrl+F for help in this FAQs.md.

Contributors

I would like to express my sincere gratitude for the contributions made by all the contributors.


License

MIT
I hold no legal responsibility;

The MIT License (MIT)

Copyright (c) 2023 Minwoo Park

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Bugs and Issues

Sincerely grateful for any reports on new features or bugs. Your valuable feedback on the code is highly appreciated.

Contacts

Reference

[1] https://huggingface.co/docs/api-inference/index


Copyright (c) 2023 MinWoo Park, South Korea

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

transllm-0.1.8.tar.gz (7.2 kB view details)

Uploaded Source

Built Distribution

transllm-0.1.8-py3-none-any.whl (7.8 kB view details)

Uploaded Python 3

File details

Details for the file transllm-0.1.8.tar.gz.

File metadata

  • Download URL: transllm-0.1.8.tar.gz
  • Upload date:
  • Size: 7.2 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.9.12

File hashes

Hashes for transllm-0.1.8.tar.gz
Algorithm Hash digest
SHA256 b97bb6ffa3b2204795f7a9a0a41e175033223d06840d6a2341942b693075977e
MD5 e855f09cce4a51e2d3f098253a76924c
BLAKE2b-256 9ec5f3768e06837a3eb8bfe6d98c4fc4dccc31619859bd9607c2f03ecf8d6f33

See more details on using hashes here.

File details

Details for the file transllm-0.1.8-py3-none-any.whl.

File metadata

  • Download URL: transllm-0.1.8-py3-none-any.whl
  • Upload date:
  • Size: 7.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.9.12

File hashes

Hashes for transllm-0.1.8-py3-none-any.whl
Algorithm Hash digest
SHA256 ff836a6c442e305991b9dabb4596f29170952fa7ada8590389da63d204752ce9
MD5 d2da9f170354eb697d0c59cb471d12ee
BLAKE2b-256 0c591a6613777ae0656b1734932fb50c8cdbdbf16c3f4773770d93e5202413fd

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page