feat: add OpenAI generation stats

backend/config.py

@@ -574,6 +574,7 @@ ENABLE_COMMUNITY_SHARING = PersistentConfig(
     os.environ.get("ENABLE_COMMUNITY_SHARING", "True").lower() == "true",
 )
 
+
 class BannerModel(BaseModel):
     id: str
     type: str

src/lib/apis/streaming/index.ts

@@ -8,6 +8,16 @@ type TextStreamUpdate = {
 	citations?: any;
 	// eslint-disable-next-line @typescript-eslint/no-explicit-any
 	error?: any;
+	usage?: ResponseUsage;
+};
+
+type ResponseUsage = {
+	/** Including images and tools if any */
+	prompt_tokens: number;
+	/** The tokens generated */
+	completion_tokens: number;
+	/** Sum of the above two fields */
+	total_tokens: number;
 };
 
 // createOpenAITextStream takes a responseBody with a SSE response,
@@ -59,7 +69,11 @@ async function* openAIStreamToIterator(
 				continue;
 			}
 
-			yield { done: false, value: parsedData.choices?.[0]?.delta?.content ?? '' };
+			yield {
+				done: false,
+				value: parsedData.choices?.[0]?.delta?.content ?? '',
+				usage: parsedData.usage
+			};
 		} catch (e) {
 			console.error('Error extracting delta from SSE event:', e);
 		}

src/lib/components/chat/Chat.svelte

@@ -685,6 +685,12 @@
 				{
 					model: model.id,
 					stream: true,
+					stream_options:
+						model.info?.meta?.capabilities?.usage ?? false
+							? {
+									include_usage: true
+							  }
+							: undefined,
 					messages: [
 						$settings.system || (responseMessage?.userContext ?? null)
 							? {
@@ -753,9 +759,10 @@
 
 			if (res && res.ok && res.body) {
 				const textStream = await createOpenAITextStream(res.body, $settings.splitLargeChunks);
+				let lastUsage = null;
 
 				for await (const update of textStream) {
-					const { value, done, citations, error } = update;
+					const { value, done, citations, error, usage } = update;
 					if (error) {
 						await handleOpenAIError(error, null, model, responseMessage);
 						break;
@@ -771,6 +778,10 @@
 						break;
 					}
 
+					if (usage) {
+						lastUsage = usage;
+					}
+
 					if (citations) {
 						responseMessage.citations = citations;
 						continue;
@@ -804,6 +815,10 @@
 					}
 				}
 
+				if (lastUsage) {
+					responseMessage.info = { ...lastUsage, openai: true };
+				}
+
 				if ($chatId == _chatId) {
 					if ($settings.saveChatHistory ?? true) {
 						chat = await updateChatById(localStorage.token, _chatId, {

src/lib/components/chat/Messages/ResponseMessage.svelte

@@ -106,8 +106,13 @@
 		renderLatex();
 
 		if (message.info) {
-			tooltipInstance = tippy(`#info-${message.id}`, {
+			let tooltipContent = '';
-				content: `<span class="text-xs" id="tooltip-${message.id}">response_token/s: ${
+			if (message.info.openai) {
+				tooltipContent = `prompt_tokens: ${message.info.prompt_tokens ?? 'N/A'}<br/>
+													completion_tokens: ${message.info.completion_tokens ?? 'N/A'}<br/>
+													total_tokens: ${message.info.total_tokens ?? 'N/A'}`;
+			} else {
+				tooltipContent = `response_token/s: ${
 					`${
 						Math.round(
 							((message.info.eval_count ?? 0) / (message.info.eval_duration / 1000000000)) * 100
@@ -137,9 +142,10 @@
                     eval_duration: ${
 											Math.round(((message.info.eval_duration ?? 0) / 1000000) * 100) / 100 ?? 'N/A'
 										}ms<br/>
-                    approximate_total: ${approximateToHumanReadable(
+                    approximate_total: ${approximateToHumanReadable(message.info.total_duration)}`;
-											message.info.total_duration
+			}
-										)}</span>`,
+			tooltipInstance = tippy(`#info-${message.id}`, {
+				content: `<span class="text-xs" id="tooltip-${message.id}">${tooltipContent}</span>`,
 				allowHTML: true
 			});
 		}

src/routes/(app)/workspace/models/create/+page.svelte

@@ -56,6 +56,20 @@
 		id = name.replace(/\s+/g, '-').toLowerCase();
 	}
 
+	let baseModel = null;
+	$: {
+		baseModel = $models.find((m) => m.id === info.base_model_id);
+		console.log(baseModel);
+		if (baseModel) {
+			if (baseModel.owned_by === 'openai') {
+				capabilities.usage = baseModel.info?.meta?.capabilities?.usage ?? false;
+			} else {
+				delete capabilities.usage;
+			}
+			capabilities = capabilities;
+		}
+	}
+
 	const submitHandler = async () => {
 		loading = true;
 

src/routes/(app)/workspace/models/edit/+page.svelte

@@ -107,6 +107,10 @@
 				params = { ...params, ...model?.info?.params };
 				params.stop = params?.stop ? (params?.stop ?? []).join(',') : null;
 
+				if (model?.owned_by === 'openai') {
+					capabilities.usage = false;
+				}
+
 				if (model?.info?.meta?.capabilities) {
 					capabilities = { ...capabilities, ...model?.info?.meta?.capabilities };
 				}