ConardLi/easy-dataset 1.3.7

[1.3.7] 2025-06-11

on GitHub

如果遇到 Github 下载慢的问题可以使用网盘下载：https://pan.quark.cn/s/194b7eedf16e

🔧 修复

1. 视觉模型PDF处理客户端报错
   → 解决视觉模型解析PDF时在客户端环境的兼容性报错，确保跨平台稳定运行。
2. NPM install Canvas模块编译失败
   → 修复Canvas模块在不同系统环境下的编译异常，完善依赖安装流程。
3. 部分推理模型思维链获取失败（#381）
   → 修正推理模型输出解析逻辑，确保思维链内容完整提取至问题关联字段。
4. 批量生产GA并发数限制（#385）
   → 解除批量生成GA数据时最多同时处理10个任务的限制，支持自定义并发配置。
5. 文件列表展示数量限制（#350）
   → 修复文件列表仅显示前10条的问题，支持完整展示所有上传文件。

⚡ 优化

1. 文献处理异步化改造
   → 重构文献处理流程为后台异步任务，支持实时查看处理进度条与状态日志。
2. GA提示词污染修复
   → 清理提示词模板中的冗余字符与格式干扰，确保生成内容纯净度。
3. 模型操作前置校验
   → 未选择模型时自动禁用相关功能按钮，避免因参数缺失导致的非预期报错。
4. 新建模型提示优化
   → 新增输入提示文本，明确告知用户可自定义模型提供商（如OpenAI/本地部署）及模型名称。
5. Playground界面功能增强（#381）
   → 在交互测试界面新增思维链展示区域，实时可视化推理模型的思考过程。

🔧 Fixes

1. Visual model PDF processing error in client environment
   → Resolved compatibility errors when visual models parse PDFs in client environments for cross-platform stability.
2. NPM install Canvas module compilation failure
   → Fixed compilation issues of the Canvas module across different systems to improve dependency installation.
3. Missing thought chains in some inference models（#381）
   → Corrected the parsing logic for inference model outputs to ensure complete extraction of thought chains into question fields.
4. Batch GA generation concurrency limit（#385）
   → Removed the restriction of processing only 10 tasks simultaneously in batch GA generation, supporting custom concurrency configuration.
5. File list display limit（#350）
   → Fixed the issue where only the first 10 files were shown, enabling full display of all uploaded files.

⚡ Optimizations

1. Asynchronous literature processing
   → Refactored literature processing as background asynchronous tasks with real-time progress tracking and status logs.
2. GA prompt contamination fix
   → Cleaned redundant characters and format interference in prompt templates to ensure pure generated content.
3. Model operation pre-checks
   → Automatically disables related function buttons when no model is selected to prevent unexpected errors from missing parameters.
4. New model creation prompt enhancement
   → Added input hints to inform users that both model providers (e.g., OpenAI/local deployment) and model names can be customized.
5. Playground interface improvement（#381）
   → Added a thought chain display area in the interactive testing interface to visualize the reasoning process of inference models in real time.