fix(vlm): default max_tokens=32768 exceeds gpt-4o-mini completion cap → silent 0-memory extraction (#2751)#2755

Open

r266-tech wants to merge 3 commits into

volcengine:mainfrom

r266-tech:fix/vlm-default-max-tokens-2751

Commits on Jun 21, 2026

fix(vlm): default max_tokens fallback exceeds gpt-4o-mini completion cap (#2751)
r266-tech
committed
test(vlm): regression for default max_tokens fallback within model cap (#2751)
r266-tech
committed

Commits on Jun 23, 2026

fix(vlm): keep reasoning-model unset max_tokens at prior 32768 default
r266-tech
committed