final: sync completed submission
#3
by Dracufeuer - opened
Sync final GitHub-tested DOTA2Tuned submission state: Ask thinking toggle, profile-routed Modal inference, updated README/PLAN/MODEL_SELECTION, compact artifacts, and agent instructions.
Co-authored-by: Codex noreply@openai.com
Final project finished.
- GitHub main is pushed at
2774c11 feat(model): gate thinking mode. - Modal was redeployed and live-smoked for Tiny, Balanced, and Quality inference.
- Balanced MiniCPM Thinking is enabled with recommended
768max tokens; Tiny/Quality Qwen Instruct profiles correctly disable Thinking. - README/PLAN/MODEL_SELECTION and compact serving artifacts are included.
- Demo video link and Build Small validator tags are present in README.
Co-authored-by: Codex noreply@openai.com