From f21366e55f45ef0656dcdf2f566c64435146c8b8 Mon Sep 17 00:00:00 2001 From: Julian Freeman Date: Wed, 25 Feb 2026 00:00:07 -0400 Subject: [PATCH] improve evaluation --- src/App.vue | 108 ++++++++++++++++++++++++++++++++++++----- src/stores/settings.ts | 4 +- 2 files changed, 100 insertions(+), 12 deletions(-) diff --git a/src/App.vue b/src/App.vue index 3b89309..35bf66b 100644 --- a/src/App.vue +++ b/src/App.vue @@ -90,20 +90,23 @@ const sourceDropdownOpen = ref(false); const targetDropdownOpen = ref(false); const speakerDropdownOpen = ref(false); const toneDropdownOpen = ref(false); +const evaluationProfileDropdownOpen = ref(false); const closeAllDropdowns = () => { sourceDropdownOpen.value = false; targetDropdownOpen.value = false; speakerDropdownOpen.value = false; toneDropdownOpen.value = false; + evaluationProfileDropdownOpen.value = false; }; -const toggleDropdown = (type: 'source' | 'target' | 'speaker' | 'tone') => { +const toggleDropdown = (type: 'source' | 'target' | 'speaker' | 'tone' | 'evaluationProfile') => { const states = { source: sourceDropdownOpen, target: targetDropdownOpen, speaker: speakerDropdownOpen, - tone: toneDropdownOpen + tone: toneDropdownOpen, + evaluationProfile: evaluationProfileDropdownOpen }; const targetState = states[type]; @@ -186,6 +189,12 @@ const currentToneLabel = computed(() => { return TONE_REGISTER_OPTIONS.find(opt => opt.value === settings.toneRegister)?.label || '正式专业'; }); +const currentEvaluationProfileLabel = computed(() => { + if (!settings.evaluationProfileId) return '使用主翻译配置（默认）'; + const profile = settings.profiles.find(p => p.id === settings.evaluationProfileId); + return profile ? `${profile.name} — ${profile.modelName}` : '使用主翻译配置（默认）'; +}); + const swapLanguages = () => { const temp = { ...settings.sourceLang }; settings.sourceLang = { ...settings.targetLang }; @@ -216,6 +225,20 @@ const evaluateTranslation = async () => { isEvaluating.value = true; evaluationResult.value = null; + // Determine which API config to use for evaluation + let apiBaseUrl = settings.apiBaseUrl; + let apiKey = settings.apiKey; + let modelName = settings.modelName; + + if (settings.evaluationProfileId) { + const profile = settings.profiles.find(p => p.id === settings.evaluationProfileId); + if (profile) { + apiBaseUrl = profile.apiBaseUrl; + apiKey = profile.apiKey; + modelName = profile.modelName; + } + } + const evaluationPrompt = settings.evaluationPromptTemplate .replace(/{SOURCE_LANG}/g, sourceLang.value.englishName) .replace(/{TARGET_LANG}/g, targetLang.value.englishName) @@ -226,7 +249,7 @@ const evaluateTranslation = async () => { .replace(/{TRANSLATED_TEXT}/g, targetText.value); const requestBody = { - model: settings.modelName, + model: modelName, messages: [ { role: "system", content: "You are a professional translation auditor. You must respond in valid JSON format." }, { role: "user", content: evaluationPrompt } @@ -238,8 +261,8 @@ const evaluateTranslation = async () => { try { const response = await invoke('translate', { - apiAddress: settings.apiBaseUrl, - apiKey: settings.apiKey, + apiAddress: apiBaseUrl, + apiKey: apiKey, payload: requestBody }); @@ -616,10 +639,10 @@ const translate = async () => {

-
+
- 建议优化 + 建议优化

{{ evaluationResult.improvements }}

@@ -773,6 +796,69 @@ const translate = async () => { )">
+ +
+
+ + +
+ + +
+ + + +
+ +
+ +
+ +
+ +

+ 提示：建议为审计选择更强大的模型以获得更精准的反馈 +

+
@@ -789,6 +875,9 @@ const translate = async () => { rows="6" class="w-full px-4 py-3 border dark:border-slate-700 rounded-lg bg-transparent focus:ring-2 focus:ring-blue-500/20 focus:border-blue-500 outline-none transition-all font-mono text-xs leading-relaxed text-slate-900 dark:text-slate-100" > +
+ {{ tag }} +
@@ -802,13 +891,10 @@ const translate = async () => { class="w-full px-4 py-3 border dark:border-slate-700 rounded-lg bg-transparent focus:ring-2 focus:ring-blue-500/20 focus:border-blue-500 outline-none transition-all font-mono text-xs leading-relaxed text-slate-900 dark:text-slate-100" >
- {{ tag }} + {{ tag }}
-
- {{ tag }} -
diff --git a/src/stores/settings.ts b/src/stores/settings.ts index 675ea34..c1587ff 100644 --- a/src/stores/settings.ts +++ b/src/stores/settings.ts @@ -67,7 +67,7 @@ Your task is to critically evaluate the accuracy and quality of a translation. [Instructions] 1. Compare the [Source Text] and [Translated Text] meticulously. -2. Check if the translation respects the [Context Info] and [Speaker Identity]. +2. Check if the translation respects the [Context Info]. 3. Assign an "Accuracy Score" from 0 to 100. - Give 0 if there are fatal semantic errors, complete hallucinations, or if the meaning is reversed. - Deduct points for minor inaccuracies, unnatural phrasing, or tone mismatches. @@ -101,6 +101,7 @@ export const useSettingsStore = defineStore('settings', () => { const enableEvaluation = useLocalStorage('enable-evaluation', true); const evaluationPromptTemplate = useLocalStorage('evaluation-prompt-template', DEFAULT_EVALUATION_TEMPLATE); + const evaluationProfileId = useLocalStorage('evaluation-profile-id', null); // 存储整个对象以保持一致性 const sourceLang = useLocalStorage('source-lang-v2', LANGUAGES[0]); @@ -133,6 +134,7 @@ export const useSettingsStore = defineStore('settings', () => { systemPromptTemplate, enableEvaluation, evaluationPromptTemplate, + evaluationProfileId, sourceLang, targetLang, speakerIdentity,