iFlytek Xingchen MaaS 136

What is iFlytek Star MaaS?

iFlytek Star MaaS is a one-stop AIGC content operations platform developed by iFlytek, primarily targeted at developers. It provides large model custom training services, brings together numerous high-quality models, offers full lifecycle management of large models, features strong data engineering capabilities and a great fine-tuning experience, and is compatible with the OpenAI protocol, covering multiple scenarios such as content creation, coding, and logical reasoning.

The iFlytek Star MaaS platform encompasses a variety of functions, including model creation, model marketplace, experience center, and solutions. In terms of model processing, it supports operations such as online inference, batch inference, model fine-tuning, model distillation, model hosting, model evaluation, and model management. It also includes a data management section covering datasets and data augmentation. Additionally, there are features such as Prompt engineering, system configuration, and billing management.

iFlytek Star MaaS Features

• Rich Model Selection: Brings together over 20 well-known industry models such as iFlytek Spark, Llama3, and SD-XL, and newly adds self-developed Spark models including Spark Max, Spark Mini, and Spark Tiny. It also introduces fine-tuned versions of multiple open-source models such as the InternLM series and Qwen2.5 series to meet diverse business scenarios.

• Strong Data Engineering Capabilities: Builds comprehensive data engineering capabilities based on dimensions such as Q&A extraction, data augmentation, and prompt engineering. It supports data generalization for common text generation, semantic understanding, and knowledge Q&A, expands dataset quantity, supports mixed training, and achieves support for nearly all mainstream industry dataset formats, such as ShareGPT and Alpaca format datasets.

• Excellent Fine-Tuning Experience: Provides a flexible and ready-to-use full-stack toolchain, supporting the splitting of a certain proportion from the training set for validation, and supports up to 10 models performing online batch inference simultaneously, greatly improving data processing efficiency.

• Compatible with OpenAI Protocol: Makes it convenient for developers to migrate capabilities. Users who previously used OpenAI's capabilities can switch to Spark's capabilities with one click, achieving a "seamless" migration.

• Supports function_call Fine-Tuning Training: Enables the model to understand when to call a function, which function to call, and how to correctly pass parameters to the function, thereby enhancing the model's practicality in real-world applications and enabling better interaction with external tools or systems.

• Convenient Model Performance Evaluation: Supports Loss curve display and referee model performance evaluation comparison. During the fine-tuning modality training process, refined data ratios and dynamic adjustments can be used to enhance capabilities in the target domain while maximally avoiding catastrophic forgetting and maintaining the model's overall performance.

iFlytek Star MaaS Application Scenarios

• Content Creation: For scenarios such as high-end content creation and professional knowledge services, the Spark Max model can be used.

• Online Customer Service: For online customer service systems of small startup companies, the Spark Tiny model can be used to meet performance requirements while saving costs.

• Industry Customization: Targeting specific industries and scenarios, by "feeding" specific data to general large models, fine-tuning can easily combine the advantages of industry "specialists" while maintaining the strengths of "generalists," meeting the specific needs of the industry.

How to Use iFlytek Star MaaS

1. Register and Log In: Visit the official website of iFlytek Star MaaS, create an account, or log in with an existing account.

2. Select a Model: Choose an appropriate pre-trained model based on business needs.

3. Data Preparation: Prepare or select suitable datasets for fine-tuning; the platform supports multiple dataset formats.

4. Model Fine-Tuning: Use the tools provided by the platform to fine-tune the selected model, including steps such as uploading datasets and configuring fine-tuning parameters.

5. Model Evaluation: Use evaluation tools to assess the performance of the fine-tuned model.

6. Model Deployment: Deploy the trained model to real-world applications.

7. API Invocation: Invoke models based on the API provided by iFlytek Star to integrate model services.

Who iFlytek Star MaaS Is For

• AI Developers: Developers who need to perform deep learning model training.

• AI Researchers: Researchers who need to conduct custom model training.

• Enterprise Users: Enterprises that need customized large models to meet specific business requirements.