dsh-livebench-panel 0.2.6 → 0.2.8

package/lib/index.js

@@ -691,7 +691,12 @@ function apply(ctx) {
 			? null
 			: reasoningEffort;
 		const hasEffortSuffix = effort !== null && effort !== "off";
-		const displayName = displayModelName(providerId || "direct", modelId) + (hasEffortSuffix ? "@" + effort : "");
+		// 每次评测使用带时间戳的独立 display-name：答案/判分文件各自独立，
+		// 成绩表里每次评测是独立一行，可对比模型的历史变化（不再覆盖旧数据）。
+		const _now = new Date();
+		const _pad = (n) => String(n).padStart(2, "0");
+		const runStamp = `r${_now.getFullYear()}${_pad(_now.getMonth() + 1)}${_pad(_now.getDate())}-${_pad(_now.getHours())}${_pad(_now.getMinutes())}${_pad(_now.getSeconds())}`;
+		const displayName = displayModelName(providerId || "direct", modelId) + (hasEffortSuffix ? "@" + effort : "") + "__" + runStamp;
 		let cliModel = modelId;
 		let writeError = null;		// Anthropic-protocol proxies cannot go through --api-base (that path
 		// speaks OpenAI Chat Completions). Instead the generated model config
@@ -725,7 +730,7 @@ function apply(ctx) {
 			...(benchNames !== null ? benchNames : [benchParts.join("/")]),
 			"--livebench-release-option", release,
 			"--max-tokens", String(asInt(body.maxTokens, 256, 32768, 32000)),
-			"--parallel-requests", String(asInt(body.parallel, 1, 8, 2)),
+			"--parallel-requests", String(asInt(body.parallel, 1, 8, 1)),
 			// 流式：中转网关（Cloudflare 等）对非流式请求有 ~100s 超时（524），
 			// 高推理强度模型思考数分钟必然超时。流式保持字节流动可规避。
 			"--stream",
@@ -831,18 +836,18 @@ function apply(ctx) {
 			// 自动补跑：正常结束但存在 $ERROR$ 答案时，用 --resume --retry-failures
 			// 只重跑失败题（最多 2 轮）。统计改为异步流式（同步版曾阻塞事件循环，
 			// 且函数缺失会在 close 回调抛未捕获异常 → DSH 假死断连）。
-			if (record.exitCode === 0 && record.autoRetries < 2 && record.body) {
+			if (record.exitCode === 0 && record.autoRetries < 3 && record.body) {
 				(async () => {
 					try {
 						const errors = await countErrorAnswersAsync(layout.dataDir, record.displayName, record.body);
 						if (errors > 0) {
-							record.log.push(`[自动补跑] 检测到 ${errors} 条失败答案，自动重试（第 ${record.autoRetries + 1}/2 轮）`);
+							record.log.push(`[自动补跑] 检测到 ${errors} 条失败答案，自动重试（第 ${record.autoRetries + 1}/3 轮）`);
 							const retryBody = { ...record.body, resume: true, retryFailures: true };
 							setTimeout(() => {
 								startRun(retryBody, record.autoRetries + 1).catch((e) => {
 									appendLog(record, `[自动补跑启动失败] ${e.message}`);
 								});
-							}, 3000);
+							}, 15000);
 						}
 					} catch (e) {
 						record.log.push(`[自动补跑检查失败] ${e.message}`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-livebench-panel",
-  "version": "0.2.6",
+  "version": "0.2.8",
   "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.",
   "license": "MIT",
   "type": "module",