feat: add OpenAI generation stats

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-${message.id}`, {
-				content: `<span class="text-xs" id="tooltip-${message.id}">response_token/s: ${
+			let tooltipContent = '';
+			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(
-											message.info.total_duration
-										)}</span>`,
+                    approximate_total: ${approximateToHumanReadable(message.info.total_duration)}`;
+			}
+			tooltipInstance = tippy(`#info-${message.id}`, {
+				content: `<span class="text-xs" id="tooltip-${message.id}">${tooltipContent}</span>`,
 				allowHTML: true
 			});
 		}