neurips-ed2026-anon-checkpoints/submission-eval-artifacts

NeurIPS ED 2026 Anonymous Evaluation Artifacts

This dataset repo contains sanitized evaluation artifacts for an anonymous NeurIPS ED 2026 submission. It is metadata-focused: normalized benchmark JSON, selected small paper-facing summaries, reviewer indexes, and manifests.

Checkpoint artifacts are referenced through neurips-ed2026-anon-checkpoints/submission-checkpoints. This dataset repo does not contain model checkpoints or model weights.

Anonymous code artifact: https://anonymous.4open.science/r/lowrankarena-7883

Start Here

1. Open paper_results_index.csv to map each paper table/figure to summary files, raw JSON/result paths, checkpoint families, command references, and coverage status.
2. Open paper_item_coverage.csv for a compact present/partial/missing/manual-review view.
3. Open artifact_locations.md for repository layout and where large raw results live.
4. Use raw_results_inventory.csv to inspect the checkpoint repo eval_artifacts/results/ tree without downloading it.
5. Use paper_results/artifact_command_index.csv and paper_results/serving_e2e_summary.csv for command hints and exact serving commands where embedded.
6. Check KNOWN_LIMITATIONS.md before interpreting partial or estimate-only paper items.

Key Files

• paper_results_index.csv: current-paper coverage index for Table 1, Figures 4-7, and Tables 2-18.
• raw_results_inventory.csv: HF tree metadata for checkpoint-repo raw results.
• dataset_repo_inventory.csv: HF tree metadata for this dataset repo at index-generation time.
• paper_item_coverage.csv: compact coverage/status table.
• paper_results/result_inventory.csv: normalized JSON inventory already included in this dataset repo.
• paper_results/metric_flattened.csv: flattened metrics across normalized evaluation JSON.
• paper_results/serving_e2e_summary.csv: endpoint serving metrics and embedded vllm bench serve commands.
• paper_results/structured_pruning_quality_summary.csv: sanitized small summary used for the structured-pruning side of Figure 5.

Raw Results and Checkpoints

Large raw result folders stay in the checkpoint repo:

neurips-ed2026-anon-checkpoints/submission-checkpoints/eval_artifacts/results/

Checkpoint weights also stay in the checkpoint repo. The dataset repo references checkpoint families and result paths only.

Exclusions

This dataset repo intentionally excludes checkpoint directories, model weights, server logs, stdout/stderr logs, private planning files, token files, and private source maps. Files with weight-like extensions such as .safetensors, .bin, .pt, .pth, and .ckpt must not be uploaded here.

Generated at: 2026-05-07T02:33:21Z