Taizhi Cloud's AI Foundry Service (AFS) is a complete enterprise-grade generative AI solution, providing a one-stop solution from large model optimization to inference cloud/ground deployment. It includes "AFS Platform + AFS Cloud" to help enterprises fine-tune their own models and build inference services, and "AFS ModelSpace" which includes a variety of enhanced Formosa Models (FFM) with Traditional Chinese support and open-source trusted large models to help enterprises quickly perform cloud inference. All of the above services can be operated using the No-code platform.

If users have a need to build enterprise-specific models, they can quickly select the FFM base model and high-performance throughput computing power recipe through the "AFS Platform" to complete the fine-tuning training of hundreds of millions of tokens in a few hours. At the same time, by using "AFS Cloud" to automatically create inference service endpoints and independent API URLs, they can carry out enterprise generative AI application integration and Chat/Playground testing.

For enterprises that do not need further model fine-tuning, they can choose the latest FFM or open-source large models on "AFS ModelSpace" and directly call the API or perform inference tasks through Chat/Playground. "AFS ModelSpace" also offers different billing models, using tokens or models by GPU-hours.

1. "AFS Platform + AFS Cloud": Offers a billing model based on "model creation time," with costs varying depending on the selected base model.
2. "AFS ModelSpace": Offers a dual-mode billing system based on "number of tokens used" and "model creation time". The cost will vary depending on the base model selected.

For more details, please seePrice Information.

1. For users who need to build enterprise-specific large models, the "AFS Platform + AFS Cloud" can be used to help enterprises fine-tune their specific models and build inference services.
2. For users who do not require fine-tuning of the model, "AFS ModelSpace" can be used, which allows real-time interaction with various FFM and open-source large models.
3. Users with sensitive data and cybersecurity requirements can choose the "FFM Site Licensing Solution," which deploys a dedicated large-scale site in an enterprise data center/private cloud.

 For more details, please see AFS Product Solution.

Taizhi Cloud's AI Foundry Service (AFS) is tailored for enterprise users, meeting their needs in areas such as cybersecurity, SLAs, maintenance services, and compliance. AFS offers a complete end-to-end service, including POC validation for enterprise users, customized model optimization (fine-tuning), and model deployment and inference in both on-premises and cloud-based hybrid modes. Users can choose the appropriate service based on their actual needs. Rest assured, Taizhi Cloud will not retain any customer data or records related to the use of AFS services or interactions with any products, nor will it use these data as training material for FFM.

Furthermore, the entire AFS service suite is built within non-government Class A computing facilities, compliant with national information security laws. User services and data are subject to the highest level of information security requirements. For even more stringent compliance needs, Taizhi Cloud's AI Foundry Service (AFS) is also available, tailored specifically for enterprise users, meeting their requirements for information security, SLAs, maintenance services, and compliance. AFS offers a complete end-to-end service, including enterprise POC verification, customized model optimization (fine-tuning), and hybrid model deployment and inference in both on-premises and cloud environments. Users can choose the appropriate service based on their specific needs. Rest assured, Taizhi Cloud will not retain any customer data or records related to the use of the entire AFS service suite or interactions with any products, nor will it use these data as training material for FFM.

Furthermore, the entire AFS service suite is built on non-government Class A computing facilities, in accordance with the National Information and Communications Security Act, ensuring that user services and data are subject to the highest level of information security requirements. For even more sensitive requirements, Taizhi Cloud's "FFM Ground Licensing Solution" can be selected to directly deploy a dedicated model in a ground environment, ensuring that internal enterprise data is not leaked.

If the amount of internal and external data specific to a company's domain reaches as high as 1 billion tokens, it is recommended to use a two-stage optimization training process of Pretrain-SFT to build the best-performing company-specific domain model. Adding a new dictionary ensures that the company's domain knowledge includes patent translations and a dedicated thesaurus. For example, a specific word must correspond to a specific translation or domain-specific meaning. Therefore, building a dedicated dictionary is equivalent to training the model with this new dictionary from the pretraining stage.

If there is insufficient data for pretraining, SFT can be performed directly. SFT requires at least 270,000 (partially efficient parameter fine-tuning) to 570,000 (full parameter fine-tuning) tokens. 

The QA Pair's question and answer collection is in JSON format, and the total number of tokens for each document does not exceed 4096.

The Formosa Foundation Model (FFM) is a large-scale language model for Traditional Chinese developed by Taizhiyun based on Large Language Model (LLM) technology. Taizhiyun's in-house R&D team, drawing on years of experience in the NLP field, further enhances the model's performance by strengthening the pre-trained model with LLM technology. It exhibits significantly higher quality in understanding Traditional Chinese semantics and knowledge domains, while maintaining the performance capabilities of the original model. Specifically designed for the diverse application needs of Taiwanese enterprise users across various fields, FFM also demonstrates high-quality text content generation results, possessing global knowledge and multilingual capabilities.

For more details, please see Formosa Foundation Models.

The FFM model series, from BLOOM, Llama2, and Mistral to the latest Llama3, offers enhanced Traditional Chinese versions of various open-source models. Regarding model specifications, it provides the most complete parameter specifications for each model, including large-scale models with massive parameters such as BLOOM-176B, Mixtral-8x7B, and Llama3-70B. A Phi-3 version of FFM is also under development; please stay tuned.

In the face of rapidly changing LLM application needs, Taizhi Cloud will be committed to providing the latest, most complete, diverse and high-quality large language model specifications and application services to accelerate the implementation of AI 2.0 in enterprises.

For more details, please see Formosa Foundation Models.

It is supported. This can be achieved through LangChain's customization options (see the following link:).https://python.langchain.com/en/latest/modules/models/llms/examples/custom_llm.htmlApplication system developers can write a custom LLM wrapper based on the manual and specify the address where the user deploys the FFM large language model.

Welcome to register as a member of Taizhi Cloud. You can activate the "Trial Project Application" to get a NT$10,000 credit limit for 60 days of free use and quickly experience "AFS ModelSpace" and other cloud platform products.

For more details, please seeFree Trial Project DescriptionFor personalized service, please contact us. sales@twsc.io orContact Us.

OpenAI's GPTs and its GPT marketplace/store primarily serve general users, aiming to provide GPT services for personal use. In contrast, Taizhi Cloud mainly provides enterprise-level users with reliable and controllable open-source large language model solutions, while also offering on-premises inference services that meet enterprise compliance, security, and auditing requirements.

The main differences are as follows:

1. Nocode Platform – Taizhi Cloud provides zero-code model training, fine-tuning, and inference services.
   AFS is a complete no-code platform that is easy and quick to get started with, from model training to inference deployment; while NVIDIA provides the development framework NeMo, as well as DGX Cloud, which requires writing jobs and related programs.
2. Trustworthy Open Source Large Language Model – Taiwan Intelligence provides a trustworthy open source large language model solution that incorporates Taiwan local knowledge and enhances Traditional Chinese corpora.
   FFM is a large language model trained with local Taiwanese knowledge and enhanced Traditional Chinese corpus; while the model provided by NVIDIA is a pretrained model that mainly supports English tasks and has not been enhanced with Taiwanese knowledge and Traditional Chinese corpus.
3. Billing Models – Taizhi Cloud offers different billing models for model fine-tuning and inference service deployment, depending on the selected service.
   Using the NVIDIA AI Foundry Service requires paying for the NVIDIA AI Enterprise Software License and the infrastructure costs associated with cloud deployment. Furthermore, using DGX Cloud incurs a hefty monthly usage fee.