2025.3.24

Smart Vision Ver3.4 have been released, with the following key upgrades:

• Application

1.AI application: Uploading files in Agent Application can seamlessly integrate with the workflow, achieving intelligent flow and processing of documents/images/audio and video

2.Workflow: The Start node supports batch uploading of files, intelligent splitting and parallel processing, achieving synchronous completion of multiple tasks

3.Workflow: Add List Operator node to precisely control the processing flow of different files through filtering and extraction

4.Workflow: Added a file upload interface and integrated it into the workflow API, supporting the transfer of files into the workflow

5.Workflow: For running workflows, real-time termination intervention is supported to achieve precise blocking of task processes

• Data and Capability

1.Knowledge: Added CustomRecursiveTextSplitter for chunk settings, supporting users to independently choose segmentation identifiers

2.Datasets: For the generated dataset, add the data generate type used for display and the data resources used for generation

3.Datasets: For the dataset being generated, manual termination of the generation task and restart of data generation are supported

4.Datasets: For datasets that fail to generate, add display of dataset generation task status and dashboard details

• Model

1.Model Store: 2 large models are available, including Alibaba-QwQ-32B and Google-Gemma-3-27B-IT, providing more choices for enterprise application scenarios

2.Model Training: For models training failed, clicking "Retraining" supports adjusting parameters based on the failed task

3.Model Training: Optimize model training interaction, add more container scheduling related log information to training logs