Model Card: TranslationGPT-1.2

Model Overview

TranslationGPT-1.2 is an 8B-parameter large language model checkpoint for Chinese ↔ English patent text translation, optimized for China–US patent drafting and prosecution style. It is distributed as open weights for use with standard inference stacks (e.g. vLLM) and the Hiro Translation API gateway.

Field	Value
Model name	TranslationGPT-1.2
Parameters	~8B
Languages	Chinese (zh), English (en)
Domains	Patent specifications, claims, and related technical text
Directions	zh2en, en2zh
Weights	Hugging Face: PatSnap/TranslationGPT-1.2

Intended Use

• Research, development, and engineering evaluation of patent-oriented machine translation
• Productivity assistance for drafting or reviewing patent-related Chinese/English text
• Deployment behind the Hiro Translation gateway (streaming segmentation, caching, optional QA flags)

Out-of-Scope Use

• Legal, patent prosecution, or agency advice
• Official or certified translations for filing, litigation, licensing, or regulatory submission without human review
• Processing data you are not authorized to use (e.g. unpublished applications, client-confidential materials, third-party corpora without permission)

Limitations

• Outputs may be inaccurate, incomplete, or inconsistent; human review is required before any professional or legal reliance
• Performance depends on text type, terminology, and domain; public benchmarks on proprietary patent data are summarized in the project repository
• The checkpoint does not ship with proprietary training data, client documents, or internal evaluation sets

Deployment Notes

• Recommended served name for vLLM/OpenAI-compatible APIs: translationgpt_v1_2 (must match gateway configuration)
• For installation, configuration, API reference, and benchmark tables, see the project repository below

More Information

Full documentation, gateway setup, evaluation methodology, and Disclaimer text:

https://github.com/patsnap/Hiro-Translation

Related files in that repository:

• DISCLAIMER.md — full disclaimer and terms of use
• NOTICE.md — copyright and trademark notice
• README.md — service architecture and quick start

Disclaimer

This project is provided for research, development, engineering evaluation, productivity, and commercial use purposes. It is designed to assist with Chinese-English translation of patent-related documents, but it does not provide legal advice, patent advice, patent agency services, official patent translations, or professional translation services. For the avoidance of doubt, commercial use is permitted only to the extent allowed under the applicable license terms for the source code, model weights, and documentation.

Translations or outputs generated by this software or associated model weights may be inaccurate, incomplete, inconsistent, or unsuitable for legal, technical, commercial, regulatory, or professional use. Users are solely responsible for reviewing and validating all inputs and outputs before relying on them for patent filing, prosecution, litigation, licensing, transactions, regulatory submissions, business decisions, or any other professional purpose.

Users must ensure that they have all necessary rights, permissions, and legal bases to process any documents or data with this project. Do not use this project to process confidential information, trade secrets, personal information, unpublished patent applications, invention disclosures, client materials, translation memories, terminology databases, or third-party materials unless you are authorized to do so and have assessed all applicable legal, privacy, data security, confidentiality, export control, sanctions, intellectual property, and contractual obligations.

Users are responsible for ensuring that their access to, download, deployment, modification, redistribution, and use of this software and any associated model weights comply with applicable laws and regulations, including export control, sanctions, technology import and export, data protection, cybersecurity, intellectual property, and sector-specific requirements.

This project does not include, and no rights are granted to, any proprietary patent corpus, benchmark dataset, training dataset, evaluation dataset, translation memory, terminology database, confidential client document, unpublished patent application, invention disclosure, or other third-party or proprietary material, unless expressly stated otherwise.

This project is provided “AS IS” and “AS AVAILABLE”, without warranties of any kind. To the maximum extent permitted by applicable law, the maintainers and contributors shall not be liable for any claims, damages, losses, liabilities, or other consequences arising from the use of this software, associated model weights, or any outputs generated by them.

For the complete disclaimer, see DISCLAIMER.md in the Hiro-Translation repository.