Skip to content

v0.3.90

Choose a tag to compare

@github-actions github-actions released this 08 Nov 03:52
· 5 commits to main since this release

English

  • If you have any questions or suggestions about the software, please reach us through the QQ group or GitHub Issues and we will get back to you as soon as possible.

Added

  • Two ways to measure output speed: the Usage overview and the request details now share one speed mode setting, switchable between "after the first token" and "total latency". After the first token counts only the generation phase, which tracks how fast the model actually writes; total latency includes the wait for the first token, which tracks what the request feels like end to end. The overview TPS card and its tooltip follow the selected mode, and the choice is remembered across sessions.
  • Import model capabilities from external catalogs: agent configuration can now pull model capability metadata from an external catalog, so supported options and capability flags stay current with it.

Improved

  • Speed samples and eligibility: short replies are no longer dropped from the speed figures — they report a value too. Only failed, canceled, empty-output and untimed records are excluded, and the default mode is now the first-token-subtracted one.
  • Overview TPS explanation: the tooltip now describes the calculation and the sample conditions for whichever mode is active, in Chinese, English, Japanese and Russian.

Fixed

  • Missing output limit on manual zcode model rules: hand-written model rules now get the required maxOutputTokens, so saving no longer produces calls that fail on the missing field.
  • Sticky event table header corners: removed the stray rounding so the pinned header no longer shows split white corners while scrolling.
  • Russian TPS tooltip mismatched the formula: corrected the description of the calculation baseline in Russian.

简体中文

  • 关于软件,如果您有任何问题或建议,请通过 QQ群 或 GitHub Issues 反馈,我们会尽快处理。

新增

  • 输出速度支持两种计算口径:使用记录的总览页与请求明细现在共用同一个「输出速度计算方式」选项,可在「按首字后窗口」和「按整段耗时」之间切换。按首字后窗口只统计首字之后的生成阶段,更贴近模型实际吐字速度;按整段耗时则把首字等待一并计入,更贴近用户真实体感。切换后概览页的 TPS 卡片与悬停说明同步变化,选择的模式在下次打开时保持。
  • 外部模型目录导入:智能体配置支持从外部目录导入模型能力信息,模型的可选参数与能力标记可以直接随目录更新。

优化

  • 速度统计口径与样本筛选:输出速度不再因为回复较短而被整体丢弃,短回复同样会给出速度值;仅剔除失败、取消、无输出或计时缺失的记录。默认口径调整为扣除首字延迟。
  • 总览页 TPS 说明文案:悬停提示按当前口径分别说明计算方式与有效样本条件,中、英、日、俄四种语言同步更新。

修复

  • zcode 手动模型规则缺少输出上限:为手动配置的模型规则补齐必填的 maxOutputTokens,避免保存后调用因缺少该字段而失败。
  • 固定表头的事件表格圆角:移除吸顶表头上多余的圆角,滚动时不再出现割裂的白角。
  • 俄语 TPS 说明与实际算法不符:修正俄语悬停文案中对计算基准的描述。