feat: add 48 benchmark blueprints from automated pipeline#20
Merged
Conversation
Blueprints auto-generated from published AI benchmark papers using the benchmark-pipeline (arXiv discovery → Gemini analysis → Blueprint YAML). Benchmarks include: HellaSwag, BBH, MMLU-Pro, JailbreakBench, SG-Bench, OlympicMath, MetaBench, and 41 others covering reasoning, safety, multilingual evaluation, and code generation.
… separately hellaswag.yml now contains the original HellaSwag paper (1905.07830, 10 prompts) instead of the validity critique paper (2504.07825, 4 prompts). The critique is now hellaswag-validity-critique.yml.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
benchmark-pipelinetool (arXiv discovery → PDF download → Gemini analysis → Blueprint YAML generation)blueprints/benchmarks/subdirectoryBenchmarks included
Reasoning & Knowledge:
HellaSwag, HellaSwag-Pro, BBH (BIG-Bench Hard), MMLU-Pro, MMLU-ProX, Turkish MMLU, ConceptMath, OlymMath, MR-GSM8K, STEM-PoM
Safety & Red-teaming:
JailbreakBench, SG-Bench, MART, Code-Switching Red-Teaming, Latent Jailbreak, Red-Teaming GPT-4V, MTSA, SAGE-RT
Bias & Fairness:
AI Gender Bias, HateCheckHIn, HASOC 2020/2021, Generative AI Hate Speech Detection
Code & Tool Use:
PPTC (PowerPoint Task Completion), PPTC-R
Meta-evaluation:
MetaBench, Open LLM Leaderboard, LLM-as-a-Judge, LLM-SRBench, Robustness & Reliability
Other:
BAMBOO, FFT, V-STaR, Video-MMLU, EU20, DSL, AFT Reasoning, JETTS, and more
Source papers
53 papers discovered via arXiv API, 49 successfully analyzed via Gemini, 48 Blueprints generated (4 papers had insufficient methodology for extraction).
Test plan
validate_blueprints.py