dsh-livebench-panel 0.2.15 → 0.2.16

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/lib/client.js CHANGED
@@ -539,10 +539,10 @@ window.__ModuleLoader__.load({
539
539
  h("input", { className: c("input"), type: "number", min: 256, max: 32768, value: sel.maxTokens, onChange: setField("maxTokens") }),
540
540
  ),
541
541
  h("div", { className: c("field") },
542
- h("label", { className: c("label") }, "题目序号范围(可选)"),
542
+ h("label", { className: c("label") }, "题目序号范围(可选,0 起,含首尾)"),
543
543
  h("div", { className: c("row"), style: { flexWrap: "nowrap" } },
544
- h("input", { className: c("input"), type: "number", min: 0, placeholder: "", value: sel.begin, onChange: setField("begin"), style: { flex: "1", minWidth: "0" } }),
545
- h("input", { className: c("input"), type: "number", min: 0, placeholder: "", value: sel.end, onChange: setField("end"), style: { flex: "1", minWidth: "0" } }),
544
+ h("input", { className: c("input"), type: "number", min: 0, placeholder: "起(含)", title: "起始题号(0 起,包含该题)", value: sel.begin, onChange: setField("begin"), style: { flex: "1", minWidth: "0" } }),
545
+ h("input", { className: c("input"), type: "number", min: 0, placeholder: "止(含)", title: "结束题号(包含该题);只填起不填止 = 从该题跑到底", value: sel.end, onChange: setField("end"), style: { flex: "1", minWidth: "0" } }),
546
546
  ),
547
547
  h("div", { className: c("field") },
548
548
  h("label", { className: c("label") }, "分类(可多选,不选 = 全部分类;括号内为该 release 下有效题数)"),
@@ -756,7 +756,7 @@ window.__ModuleLoader__.load({
756
756
  h("tr", null, h("th", null, "分类 / 任务"), h("td", null,
757
757
  "六大类共 18 个任务:coding(代码生成/补全)、math(竞赛数学等)、reasoning(空间推理/逻辑谜题)、language(拼写/连线/语义)、data_analysis(表格操作)、instruction_following(指令遵循)。首次验证推荐 language → typos。")),
758
758
  h("tr", null, h("th", null, "题目序号范围"), h("td", null,
759
- "从 0 起。冒烟测试填 0 到 2(只跑 3 题);2 题得分噪声很大(对一题就是 0↔100 的波动),想看真实水平建议 20 题以上。")),
759
+ "从 0 起,起止都包含。冒烟测试填 0 到 2(跑 3 题);只填「起」= 从该题跑到底。2 题得分噪声很大(对一题就是 0↔100 的波动),想看真实水平建议 20 题以上。")),
760
760
  h("tr", null, h("th", null, "max-tokens"), h("td", null,
761
761
  "单题回答的 token 上限,默认 32000。推理模型思考也占 token,不要低于 8192,否则思考被截断、答案为空会记 0 分。")),
762
762
  ),
package/lib/index.js CHANGED
@@ -834,7 +834,9 @@ function apply(ctx) {
834
834
  args.push("--question-begin", String(asInt(body.begin, 0, 100000, 0)));
835
835
  }
836
836
  if (body.end !== undefined && body.end !== null && `${body.end}`.length > 0) {
837
- args.push("--question-end", String(asInt(body.end, 0, 100000, 0)));
837
+ // 面板的「止」按用户直觉是闭区间;LiveBench 的 --question-end 是开区间
838
+ // (gen_api_answer.py: questions[begin:end]),这里 +1 对齐,否则实际少跑一题。
839
+ args.push("--question-end", String(asInt(body.end, 0, 100000, 0) + 1));
838
840
  }
839
841
  if (body.resume === true) args.push("--resume");
840
842
  if (body.retryFailures === true) args.push("--retry-failures");
@@ -884,8 +886,9 @@ function apply(ctx) {
884
886
  }
885
887
  }
886
888
 
887
- // 记录本次评测的配置范围:/results 用它计算每个任务的"设定题数"
888
- // (未做/失败的题数 = 设定题数 - 实际产出的有效答案数)。
889
+ // 记录本次评测的配置范围:/results 用它计算每个任务的"选择题数"
890
+ // 注意 begin/end 存的是面板上的原始输入(闭区间,含首尾),
891
+ // 供 computeResults 算 end - begin + 1;传给 LiveBench 时 end 已 +1。
889
892
  try {
890
893
  const metaDir = join(layout.dataDir, ".dsh_runs");
891
894
  if (!existsSync(metaDir)) mkdirSync(metaDir, { recursive: true });
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "dsh-livebench-panel",
3
- "version": "0.2.15",
4
- "description": "DSH web plugin: a LiveBench tab in the Trajectory view (right of 对话/轨迹). Run LiveBench evaluations against every model configured in the DeepSeek Harness pick provider/model, category, task, release and question range from dropdowns, watch progress, and read scores in place.",
3
+ "version": "0.2.16",
4
+ "description": "DSH web plugin: a LiveBench tab in the Trajectory view (right of ??/??). Run LiveBench evaluations against every model configured in the DeepSeek Harness ? pick provider/model, category, task, release and question range from dropdowns, watch progress, and read scores in place.",
5
5
  "license": "MIT",
6
6
  "type": "module",
7
7
  "main": "lib/index.js",