Super 3.5 evals update 20260813

[bxyu-nvidia]

What does this PR do?

1. Split Super serving prefill/decode args: benchmarks/nemotron_3.5_super/vllm_configs/nemotron_3.5_super.sh, benchmarks/nemotron_3.5_super/sbatch_external_vllm.sh
2. Upgrade to Tomer's vLLM 0.27.1 container build script: benchmarks/nemotron_3.5_super/build-super-vl-evals-v0271-thin.sh, benchmarks/nemotron_3.5_super/build-super-vl-rl-v0251-thin.sh
3. No longer downgrade to Python 3.12 since we don't need to match vLLM Python anymore due to not using Ray: benchmarks/nemotron_3.5_super/build_eval_container.sh, benchmarks/nemotron_3.5_super/downgrade_python.py
4. Cleanup sbatch script: benchmarks/nemotron_3.5_super/sbatch_external_vllm.sh
5. AA-Omniscience benchmark uses HF token in prepare: benchmarks/omniscience/prepare.py
6. Upgrade intermediate results aggregation script: scripts/print_aggregate_results.py
7. Upgrade SWE Bench resources server and OpenCodeSandboxedAgent to use PTY: resources_servers/swebench/app.py, responses_api_agents/opencode_sandboxed_agent/app.py

Checklist

• I have read the contributing guidelines.
• The change is focused; unrelated "drive-by" edits are tracked as separate issues/PRs.
• Tests added or updated and pass locally, or N/A for docs-only / non-code changes (so CI unit/server checks pass when applicable).
• Pre-commit checks pass locally (pre-commit run --all-files) (so CI lint/format/copyright pass).
• All commits have DCO sign-off (git commit -s) (so the DCO check passes).

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.