Xunfei Xingchen-HKUST Xunfei pushed AI model customization training platform | AI tool set

What is Xunfei Xingchen

Xunfei Xingchen is a customized training platform for AI models pushed by HKUST Xunfei to create your exclusive customized model. Xunfei Xingchen brings together more than 20 well -known high -quality models in the industry, such asStar Fire ModelLLAMA3, etc., support zero code fine -tuning, reduce the threshold for fine -tuning in large models. Xunfei Xingchen focuses on data management, model fine -tuning, evaluation, hosting, and reasoning services, providing large models of full life cycle management, covering multi -scenario applications such as content creation, code, and logical reasoning.

Xunfei Xingchen has completed a new upgrade in the richness of models and data sets and functional characteristics, enhances data engineering capabilities, supports multiple data sets, provides advanced parameter configuration, and meets the needs of professional developers.

Xunfei Xingchen’s main function

• Model fine -tuning deployment
  • Model integration: Provide well-known high-quality models in the 20+ industry, such as Xunxinghuo, LLAMA3, SD-XL, etc.
  • Open source model refinement: Introduce the refinement of multiple open source models such as scholar series, Qwen2.5 series.
  • New model: New Spark MAX, Spark Mini, Spark Tiny and other star fire self -developed models to meet the needs of different business scenarios.

• Data set construction and management
  • Data engineering ability: Starting from the dimensions of Q & A, data enhancement, Prompt project, etc., to build a full range of data engineering capabilities.
  • Data set format support: Support a variety of data sets, including Sharegpt, Alpaca, etc., to achieve support of mainstream data sets.
  • Automatic split test set: Platform support to automatically split a certain percentage of data from the training concentration as a test set. After the training is completed, it is verified to show the fine -tuning effect.
  • Batch reasoning: Support up to 10 models at the same time online batch reasoning, which greatly improves the efficiency of data processing, and is suitable for scenarios that need to process a large amount of data.

• Functional upgrade
  • Compatibility and migration support：Compatible with the OpenAI protocol, companies that rely on the OpenAI API can seamlessly migrate to a large domestic model and achieve a technical transition.
  • Functional enhancement and expansion: Function_call fine training,Call external tools to enhance the ability of large language models.
  • Model training and evaluation:supportDisplay the LOSS curve, the training effect of real -time monitoring model, and dynamically adjust the training strategy. The referee scoring model judging model effects。
  • Advanced parameter configuration: Support the word segmentation length, LoRa Merge, UNSLOTH and other characteristic configurations.
• API matrix：Build a rich and cost -effective API matrix to provide different versions of API services, such as Spark Pro, Ultra, and Max versions, supporting contexts and BATCH version APIs of different lengths.

How to use Xunfei Xingchen

• Register and log in: Visit XixingchenOfficial websiteEssence Create an account or log in with existing accounts.
• Understand the platform service: Browse the services and functions provided by the platform to understand the characteristics of different models and applicable scenarios.
• Choose a model: Select the right pre -training model according to business needs. Xunfei Xingchen provides a variety of models, including the Starfire Series, LLAMA3, etc.
• Data preparation: Prepare or select suitable datasets for fine -tuning. The platform supports a variety of data sets, providing tools such as data enhancement to help build high -quality data sets.
• Model fine -tuning: Fix the selected model with the provided tools. Including uploading data sets, configuration of fine -tuning parameters, etc.
• Model assessment: Use the evaluation tool to evaluate the effect of fine -tuning models, such as viewing the LOSS curve and comparing effects with the referee model.
• Model deployment: Deploy the trained model into the actual application. Xunfei Xingchen supports model hosting and reasoning services to facilitate the rapid deployment of the model.
• API call: If needed, the model calls based on the API provided by Xunfei Xingchen to achieve the integration of model services.

Xunfei Xingchen’s application scenario

• Logical reasoning: Build a smart question and answer system based on logical reasoning, provide decision support to help users quickly obtain information and solutions.
• Data management: Through automated data marking and cleaning, improve the efficiency and quality of data preparation, and provide a solid data foundation for AI model training.
• Model fine -tuning and evaluation: Customize the AI ​​model according to different business needs, and optimize the performance of the model through fine -tuning and evaluation, and adapt to specific application scenarios.
• Host and reasoning services: Deploy the trained AI model to the cloud or marginal device to provide stable and reliable online reasoning services and support high concurrency requests.
• Multiminomotical application: Training and constructing multi -modal applications. In the field of image and voice recognition, intelligent analysis and processing capabilities are provided, such as automatic recognition of image content and conversion voice as text.

© Copyright statement