README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
pretty_name: EffiBench (verl efficiency eval set)
license: other
language:
  - en
task_categories:
  - text-generation
tags:
  - code
  - code-generation
  - code-efficiency
  - benchmark
  - evaluation
  - verl

EffiBench (verl efficiency eval set)

A canonical-validated subset of DONG19/EffiBench (arXiv 2402.02037; 1000 LeetCode-style Python tasks), converted to the verl rule-reward schema by verl/scripts/data/effibench.py. License: unspecified upstream (no license tag on the HF dataset or the GitHub repo) -- treat as research-only until clarified.

Row filter (854/1000 kept). The upstream test_case column is broken for a sizeable minority of rows: tree/linked-list tasks contain literal <__main__.TreeNode object at 0x...> reprs inside asserts, in-place tasks assert on never-defined variables, 14 rows have no tests at all. The filter is executable, not heuristic: a row is kept iff its own canonical solution passes its own assert suite through the actual grading harness (974 structurally usable, 854 pass). The canonical's build-machine runtime/peak-RSS are stored in extra_info.reference_runtime_ms/reference_memory_mb as fallback references.

Scoring (verl/verl/utils/reward_score/effibench.py): the extracted class Solution runs the assert suite in ONE timed firejail execution; runtime (perf_counter around the assert loop) and peak RSS (getrusage(RUSAGE_SELF)) are measured in-process and compared against the canonical solution re-measured on the same machine (cached per task). reward = 0.5*pass + 0.25*min(1, canon_rt/rt) + 0.25*min(1, canon_mem/mem). Eval recipe: verl/recipe/run_effibench_eval.sh.

Known compromises vs the official pipeline

• Memory is peak RSS, not the paper's memory_profiler time-integral (TMU/NMU); the ~20MB interpreter baseline sits on both sides of the ratio and biases it toward 1 on small tasks.
• Assert suites are binary (first failing assert aborts): no per-case fraction.
• Efficiency ratios are clipped at 1 -- beating the canonical earns full credit, not extra credit.
• Prompts use a class Solution starter skeleton reconstructed from the canonical solution (the source ships no starter-code column).