From 8b0fe6d2ac2f68cbe5d5aeb6771d55854ce0f6de Mon Sep 17 00:00:00 2001
From: Russell Ballestrini <russell.ballestrini@gmail.com>
Date: Fri, 11 Jul 2025 15:25:02 -0400
Subject: [PATCH] apply 2048 token buffer to all upstream API requests:
 translation and TTS now calculate available tokens properly

---
 src/translation.js | 11 ++++++++---
 src/tts.js         | 24 +++++++++++++++---------
 2 files changed, 23 insertions(+), 12 deletions(-)

diff --git a/src/translation.js b/src/translation.js
index bc88806..9fd7547 100644
--- a/src/translation.js
+++ b/src/translation.js
@@ -70,10 +70,15 @@ async function* sendMessageWithHistory(messageHistory) {
 		model = getSelectedModel();
 	}
 
-	const maxTokens = getSelectedModelMaxTokens();
+	// Calculate available tokens based on input length
+	const modelMaxTokens = getSelectedModelMaxTokens();
+	const inputText = messageHistory.map(msg => msg.content).join('');
+	const inputTokens = Math.ceil(inputText.length / 4); // Simple token estimation
+	const availableTokens = Math.max(100, modelMaxTokens - inputTokens - 2048); // 2048 token buffer
+	
 	console.log("Translation using endpoint:", apiUrl);
 	console.log("Translation using model:", model);
-	console.log("Translation using max_tokens:", maxTokens);
+	console.log("Translation token calculation: max=" + modelMaxTokens + ", input≈" + inputTokens + ", available≈" + availableTokens);
 
 	const response = await fetch(apiUrl, {
 		method: "POST",
@@ -82,7 +87,7 @@ async function* sendMessageWithHistory(messageHistory) {
 			model: model,
 			messages: messageHistory,
 			temperature: 0.3, // Lower temperature for more consistent translation
-			max_tokens: maxTokens,
+			max_tokens: availableTokens,
 			stream: true,
 		}),
 	});
diff --git a/src/tts.js b/src/tts.js
index 946f18f..70725a6 100644
--- a/src/tts.js
+++ b/src/tts.js
@@ -1,6 +1,6 @@
 // Text-to-speech functionality
 import { API_KEY, MODEL, setLastTTS, TTS_API_URL } from "./config.js";
-import { getSelectedModel, getSelectedModelEndpoint } from "./models.js";
+import { getSelectedModel, getSelectedModelEndpoint, getSelectedModelMaxTokens } from "./models.js";
 
 // Simple JavaScript text cleaning for TTS
 function cleanTextForTTS(text) {
@@ -129,12 +129,9 @@ export async function extractSpokenTokens(content) {
 		.replace(/\s+/g, ' ')
 		.trim();
 
-	const payload = {
-		model: getSelectedModel(),
-		messages: [
-			{
-				role: "system",
-				content: `You are extracting the main article content for text-to-speech. The input may be markdown or pre-cleaned HTML.
+	// Calculate available tokens for TTS processing
+	const modelMaxTokens = getSelectedModelMaxTokens();
+	const systemPrompt = `You are extracting the main article content for text-to-speech. The input may be markdown or pre-cleaned HTML.
 
 Your task:
 1. Find and extract the main article content (title, subtitle, body text)
@@ -145,7 +142,16 @@ Your task:
 6. Remove any remaining HTML tags but keep the text inside them
 7. Start with the article title, then the main content
 
-Return only the article text ready for TTS, nothing else.`,
+Return only the article text ready for TTS, nothing else.`;
+	const inputTokens = Math.ceil((systemPrompt + preCleanedContent).length / 4);
+	const availableTokens = Math.max(100, modelMaxTokens - inputTokens - 2048);
+
+	const payload = {
+		model: getSelectedModel(),
+		messages: [
+			{
+				role: "system",
+				content: systemPrompt,
 			},
 			{
 				role: "user",
@@ -153,7 +159,7 @@ Return only the article text ready for TTS, nothing else.`,
 			},
 		],
 		temperature: 0,
-		max_tokens: 76000,
+		max_tokens: availableTokens,
 	};
 
 	console.log("Processing page content for TTS...");