fix: Gemma4 max_new_tokens 2048 → 4096 — verbose answers still truncating

ijbo · ijbo · commit ea6bdeed195e · 2026-04-03T11:19:01.000+09:00
diff --git a/changelogs/CHANGELOG-gemma4-tokens-4096.md b/changelogs/CHANGELOG-gemma4-tokens-4096.md
@@ -0,0 +1 @@
+# Fix: Gemma4 max_new_tokens 2048 → 4096 — verbose answers still truncating
diff --git a/public/ai-worker-gemma4.js b/public/ai-worker-gemma4.js
@@ -213,7 +213,8 @@ async function generate({ userPrompt, prompt, attachments = [], context, chatHis
 
         await model.generate({
             ...inputs,
-            max_new_tokens: options.maxTokens || 2048,
+            max_new_tokens: options.maxTokens || 4096,
+
 
             do_sample: true,
             temperature: options.temperature || 0.7,

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1 @@`
	`1`	`+# Fix: Gemma4 max_new_tokens 2048 → 4096 — verbose answers still truncating`