dsh-livebench-panel 0.2.17 → 0.2.18

package/lib/index.js

@@ -791,7 +791,6 @@ function apply(ctx) {
 		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
 		// selects LiveBench's native anthropic client and the spawn env points
@@ -803,22 +802,24 @@ function apply(ctx) {
 		const isOpenAIResponsesRoute = provider
 			&& (provider.api === "openai-responses" || provider.api === "openai_responses")
 			&& provider.baseURL;
-		const needConfig = isAnthropicRoute || isOpenAIResponsesRoute || hasEffortSuffix;
-		if (needConfig) {
-			writeError = writeGeneratedModelConfig(layout, loadYaml(profileDir), {
-				displayName,
-				modelId,
-				reasoningEffort: isAnthropicRoute ? null : effort,
-				protocol: isAnthropicRoute ? "anthropic" : isOpenAIResponsesRoute ? "openai_responses" : "openai",
-			});
-			if (writeError) return { status: 500, payload: { ok: false, error: writeError } };
-			cliModel = displayName;
-		}
+		// 每次都写自己的模型配置，并且 --model 永远用 display-name。
+		// 不能图省事把裸模型名交给 LiveBench：它会用 get_model_config(裸名) 去撞它自带的
+		// 模型库（例如 "kimi-k3" 命中 moonshotai.yml），命中后就改用那份配置的
+		// provider / api_kwargs / max_tokens —— 实测会把 temperature 悄悄变成 1.0、
+		// max_tokens 变成 131072，等于用户选的 provider 被换掉。写了自己的配置，
+		// 解析结果就是确定的 {local: <modelId>} + --api-base。
+		writeError = writeGeneratedModelConfig(layout, loadYaml(profileDir), {
+			displayName,
+			modelId,
+			reasoningEffort: isAnthropicRoute ? null : effort,
+			protocol: isAnthropicRoute ? "anthropic" : isOpenAIResponsesRoute ? "openai_responses" : "openai",
+		});
+		if (writeError) return { status: 500, payload: { ok: false, error: writeError } };
 
 		const benchParts = ["live_bench", ...(category ? [category] : []), ...(task ? [task] : [])];
 		const args = [
 			"run_livebench.py",
-			"--model", cliModel,
+			"--model", displayName,
 			"--model-display-name", displayName,
 			"--bench-name",
 			...(benchNames !== null ? benchNames : [benchParts.join("/")]),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-livebench-panel",
-  "version": "0.2.17",
+  "version": "0.2.18",
   "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",
   "author": "cszr (Vithrive)",