| # IOL-AI 2026 submission — ShayanShamsi | |
| Self-provisioning script competition entry: | |
| - script.py: env diagnostics + orchestrator (writes valid submission.csv from second 0, incremental thereafter) | |
| - wheels/: offline pip bundle (vllm 0.8.5.post1 + torch 2.6 + transformers 4.51.3) | |
| - awq14b/: Qwen3-14B-AWQ (Apache-2.0) — primary model via vLLM V0 in-process | |
| - q4b/: Qwen3-4B (Apache-2.0) — transformers fallback | |
| - pipeline: no-think floor -> adaptive-budget thinking (budget-forced close) -> weak-row re-solve -> weighted vote | |
| - Answers are computed at run time from /tmp/data/test.csv only. | |