improved google gemini

This commit is contained in:
Mic 2025-08-28 21:04:00 +02:00
parent 4ef246afc3
commit 5a3ea0b380
3 changed files with 78 additions and 63 deletions

View file

@ -345,7 +345,7 @@ class MessagesArea extends HTMLElement {
}
// Check for newlines in the batch
if (tokens.includes('\n')) {
if (tokens.endsWith('\n')) {
this.flushAccumulatingMessage();
}
}

View file

@ -72,9 +72,10 @@ export class GoogleGemini {
fetchResponse = async (messages) => {
// Smart streaming: disabilita streaming per risposte piccole
const messageLength = messages.map(m => m.parts?.map(p => p.text).join('') || '').join('');
const shouldStream = false; ///this.stream && (messageLength > 500 || !this.adaptiveStreaming);
//console.log(">>>>>>>>>> Google Gemini shouldStream: " + shouldStream);
const messageLength = messages.map(m => m.parts?.map(p => p.text).join('') || '').join('').length;
console.log(">>>>>>>>>> Google Gemini messageLength: " + messageLength);
const shouldStream = this.stream && (messageLength > 200 || !this.adaptiveStreaming);
console.log(">>>>>>>>>> Google Gemini shouldStream: " + shouldStream);
try {
let google_gemini_body = {
contents:messages

View file

@ -42,16 +42,15 @@ let tokenBatch = '';
let batchTimer = null;
let timeoutTimer = null;
let lastBatchTime = 0;
let batchStartTime = 0;
// Function to send batched tokens
function sendTokenBatch(force = false, reason = 'unknown') {
if (tokenBatch && (force || tokenBatch.length >= TOKEN_BATCH_SIZE || performance.now() - lastBatchTime >= TOKEN_BATCH_DELAY)) {
console.log(`>>>>>>>>>>>>> Sending token batch (reason: ${reason}):`, tokenBatch);
postMessage({ type: 'tokenBatch', payload: { tokens: tokenBatch } });
// Reset batch state
tokenBatch = '';
lastBatchTime = performance.now();
batchStartTime = 0;
// Clear all timers
if (batchTimer) {
clearTimeout(batchTimer);
@ -67,10 +66,6 @@ function sendTokenBatch(force = false, reason = 'unknown') {
// Function to add token to batch
function addTokenToBatch(token) {
tokenBatch += token;
// Set batch start time for the first token
if (tokenBatch.length === token.length) {
batchStartTime = performance.now();
}
// Send immediately if batch is full
if (tokenBatch.length >= TOKEN_BATCH_SIZE) {
sendTokenBatch(true, 'size-limit');
@ -119,6 +114,11 @@ self.onmessage = async function(event) {
throw new Error("[ThunderAI] Google Gemini API request failed: " + response.status + " " + response.statusText + ", Detail: " + error_message + " " + errorDetail);
}
// Check if the response is streaming (SSE/chunks)
const contentType = response.headers.get('content-type') || '';
const isStreaming = contentType.includes('text/event-stream') || contentType.includes('application/x-ndjson');
if (isStreaming) {
const reader = response.body.getReader();
const decoder = new TextDecoder("utf-8");
let buffer = '';
@ -175,6 +175,20 @@ self.onmessage = async function(event) {
}
}
}
} else {
// Non-streaming: send the entire text in a single batch
try {
const responseJson = await response.json();
const text = responseJson.candidates?.[0]?.content?.parts?.[0]?.text || '';
assistantResponseAccumulator = text;
postMessage({ type: 'tokenBatch', payload: { tokens: text } });
postMessage({ type: 'tokensDone' });
conversationHistory.push({ role: 'model', parts: [{ "text": assistantResponseAccumulator }] });
assistantResponseAccumulator = '';
} catch (e) {
taLog.error("Error parsing non-streaming response: " + e);
}
}
} else if (event.data.type === 'stop') {
stopStreaming = true;
}