engineering/uncloseai.com
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

remove 2000 token hard cap for introduction messages - allow full model completion limits

Browse source
This commit is contained in:
Russell Ballestrini 2025-07-28 10:49:29 -04:00
parent eff2eee69b
commit 88dece974a
2 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
src/chat.js
View file

@ -29,7 +29,7 @@ function calculateAvailableTokens(chatHistory, maxTokens) {
buffer = Math.max(1000, Math.floor(inputTokens * 0.5));
}
const availableTokens = Math.max(100, maxTokens - inputTokens - buffer);
const availableTokens = Math.max(2000, maxTokens - inputTokens - buffer);
console.log(`Token calculation: max=${maxTokens}, input≈${inputTokens}, buffer=${buffer}, available≈${availableTokens}`);
return availableTokens;
}

4
src/ui.js
View file

@ -151,8 +151,8 @@ async function* sendMessageWithCustomHistory(messageHistory) {
// Cap to model's max completion tokens limit
const maxCompletionTokens = getSelectedModelMaxCompletionTokens();
// For intro generation, try to use at least 500 tokens even if it's a large portion of max completion
const finalAvailable = Math.min(calculatedAvailable, maxCompletionTokens, 2000); // Cap at 2000 for reasonable intro length
// For intro generation, use the full available tokens up to the model's completion limit
const finalAvailable = Math.min(calculatedAvailable, maxCompletionTokens);
console.log("UI (intro generation) using tokens:", finalAvailable, "calculated:", calculatedAvailable, "max completion:", maxCompletionTokens, "of model max:", modelMaxTokens);
const response = await fetch(apiUrl, {