dsh-recent-usage-metrics 0.2.2 → 0.2.3

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # 更新记录
 
+## 0.2.3
+
+- 中断的行现在显示它跑了多久：被手动停止的请求没有 model，总用时回退到 wall
+  （endTime - startTime），刷新后仍是同一个数。
+- 「中断」标记移到速度列。首字若已返回就照常显示实测值，总用时也照常显示——
+  缺的只是生成速度，由速度列说明原因。
+- 中断行不计入底部平均值，但标题仍数满十行。
+
 ## 0.2.2
 
 - 修复：切换模型后抽屉卡在旧的一轮。数据源的 target 曾被缓存，而会话重建

package/README.en.md

@@ -412,6 +412,19 @@ carries `ttft`, `model`, `decode`, `tps`, `output` and `wall`.
 | **Time** | **`model`** | **model time, not the Statistics view's total** |
 | Amortised | `output ÷ model` | this plugin's own figure |
 
+**Interrupted rows:**
+
+A request that was stopped never produced an `assistant/message`, so `model`, `decode` and
+`tps` are all null. That row's **total falls back to `wall`** (`endTime − startTime`) — how
+long the request actually ran before it was stopped. It is a log fact, so it survives a reload
+unchanged, which is exactly what that column is for here.
+
+The word **"interrupted" sits in the speed column**, not the first-token column: a request that
+already streamed keeps its measured first-token time and still shows its total; the only figure
+missing is the generation rate, so that is the column that says why. Interrupted rows are
+excluded from the footer average — their total is wall-clock time, and folding it into the mean
+would pull it toward a number describing neither case — but the caption still counts all ten.
+
 **Why the total is `model` and not `wall`:**
 
 The Statistics view's total column is `wall = endTime − startTime`, which **includes the step's

package/README.md

@@ -370,6 +370,16 @@ const target = uiConversation.binding(sessionId).target('stats')
 | **总用时** | **`model`** | **模型用时，不是统计页的「总用时」列** |
 | 折算速度 | `output ÷ model` | 本插件独有 |
 
+**中断的行：**
+
+被手动停止的请求没有 `assistant/message`，因此 `model`、`decode`、`tps` 全为 null。
+那一行的**总用时回退到 `wall`**（`endTime − startTime`），也就是这条请求从发出到被
+停止实际过了多久——它是日志里的事实，刷新多少次都不变，正是这个场合要的数字。
+
+**「中断」写在速度列**，不在首字列：首字若已经返回就照样显示实测值，总用时也照常显示；
+缺的只是生成速度，所以由速度列说明原因。中断行不计入底部的平均值——它的总用时是墙钟
+时间，混进均值会拉向一个既不代表完成、也不代表中断的数；但平均行的标题仍数满十行。
+
 **总用时为何取 `model` 而不是 `wall`：**
 
 统计页的「总用时」列是 `wall = endTime − startTime`，它**把工具调用也算了进去**。

package/client/client.js

@@ -880,6 +880,12 @@ window.__ModuleLoader__.load({
     function readingOfStatsStep(entry) {
       const stats = entry.metrics
       const ttftMs = numberOrNull(stats.ttft)
+      /*
+       * A step the fold marks aborted ended without a settled reply: the user stopped it, or the
+       * upstream failed mid-request. Either way there is no `messageTime`, so `model`, `decode`
+       * and `tps` are all null and the row would be entirely blank.
+       */
+      const interrupted = stats.aborted === true || stats.interrupted === true
       /*
        * The drawer's "total" is the Statistics view's MODEL time, not its wall time.
        *
@@ -888,17 +894,15 @@ window.__ModuleLoader__.load({
        * The drawer answers "how long did this request take to answer", so it reports `model` —
        * and that figure is final the moment the reply lands, which is why it must not climb
        * while tools run.
+       *
+       * An interrupted step has no `model` at all, so it falls back to `wall`. That is the
+       * figure the user actually wants there: how long the request ran before it was stopped.
+       * It is a log fact — `endTime - startTime` — so it survives a reload unchanged, unlike
+       * the live reading, which is recomputed and lost.
        */
-      const totalMs = numberOrNull(stats.model)
+      const totalMs = numberOrNull(stats.model) ?? (interrupted ? numberOrNull(stats.wall) : null)
       const tokens = numberOrNull(stats.output)
       const speed = numberOrNull(stats.tps)
-      /*
-       * A step the fold marks interrupted or aborted never produced a settled reply: a
-       * cancelled request has no first token to subtract from, so its ttft is null and the
-       * drawer used to print the same dash a not-yet-started step would. The flag lets the
-       * cell say what actually happened instead.
-       */
-      const interrupted = stats.interrupted === true || stats.aborted === true
       return {
         key: `${entry.turn}:${entry.step}`,
         at: numberOrNull(stats.startTime) ?? numberOrNull(stats.endTime),
@@ -2020,13 +2024,19 @@ window.__ModuleLoader__.load({
      * Average of the rows on screen, computed over the ones that actually carry each
      * figure (an interrupted request may have no TTFT, a placeholder no tokens). Returns
      * null for a column with no samples so the row can print a dash instead of a fake 0.
+     *
+     * Interrupted rows are excluded from every mean. Their total is wall-clock time — it
+     * includes however long the request ran before being stopped — so averaging it against
+     * model times would pull the figures toward a number that describes neither case. The
+     * row still counts toward `count`, because the caption names the list, not the sample.
      */
     function averageOf(rows) {
       if (rows.length === 0) return null
+      const measurable = rows.filter((row) => row.interrupted !== true)
       const mean = (pick) => {
         let sum = 0
         let count = 0
-        for (const row of rows) {
+        for (const row of measurable) {
           const value = pick(row)
           if (value === null || value === undefined || !Number.isFinite(value)) continue
           sum += value
@@ -2036,6 +2046,7 @@ window.__ModuleLoader__.load({
       }
       return {
         count: rows.length,
+        measured: measurable.length,
         ttftMs: mean((row) => row.ttftMs),
         speed: mean((row) => row.speed),
         tokens: mean((row) => row.tokens),
@@ -2043,8 +2054,8 @@ window.__ModuleLoader__.load({
         amortized: mean((row) => row.amortized),
         // A column is "estimated" when any sample in it was; the tilde then still means
         // the same thing it does on a single row.
-        estimated: rows.some((row) => row.estimated === true),
-        anyEstimated: rows.some((row) => row.estimated === true),
+        estimated: measurable.some((row) => row.estimated === true),
+        anyEstimated: measurable.some((row) => row.estimated === true),
       }
     }
 
@@ -2812,18 +2823,24 @@ window.__ModuleLoader__.load({
                           'span',
                           { className: 'dsh-rum-col', role: 'cell' },
                           /*
-                           * A cancelled step has no first token to report, so the column would
-                           * otherwise show the same dash as a step that never started. Say what
-                           * happened instead; it is the one figure the row can honestly give.
+                           * The first-token column reports whatever was actually measured. An
+                           * interrupted request that had already produced its first token shows
+                           * that real figure; one stopped before any token shows the dash.
                            */
-                          entry.row.interrupted === true
-                            ? t('interrupted')
-                            : secondsText(entry.row.ttftMs, t),
+                          secondsText(entry.row.ttftMs, t),
                         ),
                         h(
                           'span',
                           { className: 'dsh-rum-col dsh-rum-primary', role: 'cell' },
-                          speedText(entry.row.speed, false, t),
+                          /*
+                           * A cancelled step never settled, so it has no generation speed. The
+                           * column says what happened instead of the same dash a step that has
+                           * not started would show — the speed column is where "this produced no
+                           * rate" belongs, while first-token and total still carry real figures.
+                           */
+                          entry.row.interrupted === true
+                            ? t('interrupted')
+                            : speedText(entry.row.speed, false, t),
                         ),
                         h(
                           'span',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-recent-usage-metrics",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Geek level for the Performance & usage setting: a live reading of the newest request (first-token latency, speed, tokens, total, amortised) plus a drawer listing the last ten, all derived from the shipped Statistics view data",
   "repository": {
     "type": "git",