--- name: benchmark-onboarder-0474fc70aff4 description: Onboard AI creativity benchmarks into HELM-compliant Scenario implementations. Use when asked to onboard a benchmark, create a HELM scenario, process a paper into evaluation code, convert a dataset to HELM format, or when the user mentions "creativity benchmark", "task.py", or "scenario.py". allowed-tools: Read, Write, WebFetch, Grep, Glob, Bash, Task user-invocable: true --- # HELM Benchmark Onboarder Transform creativity benchmark datasets into HELM Scenario implementations. ## Core Principles 1. **Dataset-first** - The dataset defines what's possible; the paper provides context 2. **Skip model outputs** - We want original prompts/stimuli, not what GPT-4 generated 3. **Verify everything** - Every field must map to real data ## Resources - For HELM code patterns, see [helm-template.md](helm-template.md) - For working examples, see [examples/](examples/) - [brainteaser.py](examples/brainteaser.py) - Multiple choice with distractor answers - [analobench.py](examples/analobench.py) - Analogical reasoning MC task - [riddlesense.py](examples/riddlesense.py) - Riddle QA with CommonsenseQA format - For benchmark queue, see [benchmarks.json](benchmarks.json) - For team-accumulated learnings, see [LEARNINGS.md](LEARNINGS.md) ## Tool Usage - **WebFetch**: Read dataset documentation, READMEs, and papers - **Read**: Examine dataset files, existing scenarios, and example code - **Bash**: Load dataset samples to inspect field structure - **Glob**: Find relevant files in benchmark repos - **Task (Explore)**: For complex benchmarks, explore dataset structure first - **Write**: Generate the final scenario.py ## When to Use User says things like: - "Onboard the BRAINTEASER benchmark" - "Create a HELM scenario for this paper" - "Process this benchmark into evaluation code" ## Workflow ### Step 1: Qualify the Benchmark Before doing any work, check if this benchmark is suitable: **Ask yourself:** - Is this a creativity benchmark? - Is there a publicly available dataset? - Is there an evaluation method? (accuracy, human ratings, metrics) If the answer to any is NO, tell the user why this benchmark doesn't qualify and stop. **Focus on primary creativity tasks.** Some papers include multiple tasks or secondary experiments (e.g., MMLU baselines, general QA comparisons). Identify and implement only the core creativity evaluation tasks described in the paper. ### Handling Multi-Task Benchmarks Some papers contain multiple tasks, not all creativity-related. When onboarding: 1. **Identify the primary creativity task(s)** - Read the paper abstract/introduction to understand the main contribution 2. **Skip secondary experimental tasks** - Papers often include baseline comparisons (e.g., MMLU, general QA) that aren't creativity benchmarks 3. **When unclear, ask the user** - If you can't determine which tasks are the main creativity evaluations, ask before proceeding ### Step 2: Examine the Dataset Load a few examples and understand the structure: ```python from datasets import load_dataset ds = load_dataset("org/dataset-name", split="test") print(ds[0]) # See field names and structure ``` **Identify fields:** - **Stimulus/Question**: The input text (question, prompt, story, etc.) - **Choices**: For MC tasks, the answer options - **Answer/Label**: The correct answer or ground truth - **Metadata**: Split info, IDs, categories **Use judgment on fields:** Avoid fields that contain model-generated outputs (e.g., fields named "gpt_response" or containing experiment results). Long text fields may be legitimate inputs for evaluation tasks (e.g., stories to analyze, passages to read). When uncertain, check the paper to understand what each field represents. ### Step 3: Extract Task Instructions and Evaluation Setup Read source materials **once** to extract both what goes into the Scenario and what's needed for evaluation configuration. This avoids redundant passes through papers. #### 3a. Task Instructions (→ Scenario) **Where to look (in order):** 1. Dataset README on HuggingFace/GitHub 2. Paper Methods section 3. Paper Appendix (often has exact prompts) 4. Codebase evaluation scripts (if available) **What to capture:** - Exact prompt wording given to models (if specified) - Input formatting requirements - Any few-shot examples used **Priority:** 1. Explicit instructions from paper/README → use exactly as written, cite location 2. Standard format for task type → note "Standard MC format" in header 3. Unclear → ask user before proceeding **Critical:** Never invent or paraphrase prompts. If the paper doesn't specify exact wording, use standard formatting and note this in the header. #### 3b. Evaluation Setup (→ Companion files) While reading source materials, identify the evaluation approach. In HELM, Scenarios just load data—metrics are configured separately in RunSpecs. | Eval Type | HELM RunSpec Pattern | Additional Output | |-----------|---------------------|-------------------| | `exact_match` | `get_exact_match_metric_specs()` | None needed | | `open_ended` | `get_open_ended_generation_metric_specs()` | None (includes BLEU-1, BLEU-4, ROUGE-L, F1) | | `summarization` | `get_summarization_metric_specs()` | None needed | | `llm_judge` | Custom with Annotator | `annotator_notes.md` | | `custom` | Needs new metric implementation | `metric_notes.md` | **For standard metrics (exact_match, open_ended, summarization):** Just note the eval_type in benchmarks.json. The Scenario stays pure. **For LLM-as-judge benchmarks, extract and output to `scenarios/benchmark_name/annotator_notes.md`:** ```markdown # Annotator Requirements: BenchmarkName Source: Paper Appendix B, Section 4.2 ## Configuration for LLMAsJuryAnnotator Judge model: GPT-4-turbo Dimensions: novelty, feasibility, significance Scale: 1-5 Likert per dimension ## Judge Prompt Template Rate the following response on {dimension} from 1 to 5. Question: {QUESTION} Response: {RESPONSE} Provide your rating as a single number. ## Notes - Human correlation: 0.82 (Paper Table 3) - Authors noted position bias in judge ``` **For custom metrics, output to `scenarios/benchmark_name/metric_notes.md`:** Document what the paper measured and how, for future metric implementation. **Update benchmarks.json:** ```json { "name": "BenchmarkName", "eval_type": "open_ended|exact_match|llm_judge|custom", "notes": "any special considerations" } ``` ### Step 4: Generate the HELM Scenario Follow HELM's standard Scenario structure: ```python """ HELM Scenario: BENCHMARK_NAME Paper: [citation or URL] Code: [GitHub repo if available] Prompt format: Question: {question} A) {choice_a} B) {choice_b} C) {choice_c} D) {choice_d} Answer: Fields used: question, choices, answer Fields skipped: gpt4_response (model output) """ from datasets import load_dataset from helm.benchmark.scenarios.scenario import ( Scenario, Instance, Input, Output, Reference, CORRECT_TAG, TEST_SPLIT ) class BenchmarkScenario(Scenario): name = "benchmark_name" # lowercase, underscores description = "org/dataset-name" # data source, NOT task description tags = ["creativity", "relevant_tag"] def get_instances(self, output_path): dataset = load_dataset("org/dataset-name", split="test") instances = [] for item in dataset: # Build prompt from dataset fields prompt = f"Question: {item['question']}\n" # Build references (all choices for MC, correct answer tagged) references = [...] instances.append(Instance( input=Input(text=prompt), references=references, split=TEST_SPLIT )) return instances ``` **Note:** Evaluation configuration (metrics, annotators) is NOT part of the Scenario. Document those separately per Step 3b. **HELM conventions:** - For multiple choice: ALL choices become References, only correct one gets CORRECT_TAG - For binary (yes/no): Both options are References - For open-ended: Reference can be empty or contain gold response - `description` field = data source, not task description ### Step 5: Verify Before Finishing Check these before delivering: - [ ] Dataset loads without errors - [ ] All template variables map to real data fields - [ ] No model output fields used as inputs - [ ] References have non-empty text (for MC/closed tasks) - [ ] Correct split used (test if labels available, validation otherwise) - [ ] Test with a few examples to confirm formatting ## Output Create `scenario.py` with: - Header comment noting prompt source, fields used/skipped, paper reference - Clean, minimal code following HELM patterns ### Auto-Capture Notes to LEARNINGS.md When you encounter issues or discover patterns, append directly to LEARNINGS.md (in this skill folder): 1. Read the current LEARNINGS.md file 2. Add your note to the appropriate section (Dataset Quirks table, Common Patterns, etc.) 3. Write the updated file **Always capture notes for:** - Split issues (test has no labels, etc.) - Field name mismatches (docs say X, actual field is Y) - Special loading requirements (trust_remote_code, config names) - Skipped tasks (secondary experiments, non-creativity baselines, etc.) This ensures team knowledge is captured immediately without manual intervention. ## Common Issues | Problem | Solution | |---------|----------| | Test split has no labels | Use validation split instead (see riddlesense.py) | | Dataset requires special loading | Add `trust_remote_code=True` to load_dataset() | | Field names don't match docs | Print `ds[0]` to see actual field names | | Empty references | Wrong answer field - check the schema | | Very long prompts | Might be using wrong field (model outputs) | | Multimodal benchmark | Include if there's a text evaluation component; note modality in tags | | No explicit prompt in paper | Use standard formatting, note in header | ## Complex Benchmarks For benchmarks with multiple subsets or complex evaluation: 1. Use **Task tool with Explore agent** to understand the dataset structure first 2. Break into sub-tasks if multiple scenario files are needed 3. Use `/compact` between benchmarks if onboarding multiple in one session ## Benchmarks with LLM-as-Judge Evaluation In HELM's architecture: - **Scenarios** load data and format prompts - **Annotators** handle LLM-based judging (e.g., `LLMAsJuryAnnotator`) - **Metrics** compute final scores from annotations These are separate components. Your Scenario stays pure—evaluation config goes elsewhere. **When you encounter an LLM-as-judge benchmark:** 1. **Create the Scenario as normal** — data loading, prompt formatting 2. **References may be empty** for open-ended generation tasks—that's fine 3. **Extract annotator requirements per Step 3b** → output to `annotator_notes.md` 4. **Flag in benchmarks.json** with `"eval_type": "llm_judge"` The `annotator_notes.md` file documents what's needed to implement the `LLMAsJuryAnnotator` configuration (judge model, prompt template, dimensions, scale). This is a separate implementation task from Scenario creation. ## Examples ### Example A: Paper specifies instructions (ANALOBENCH) 1. **Qualify**: Creativity benchmark (analogical reasoning)? Yes. Dataset? Yes. Eval? Yes (accuracy). 2. **Examine dataset**: ```python ds = load_dataset("jhu-clsp/AnaloBench", "T1S1-Subset", split="train") print(ds[0].keys()) # ['Index', 'Sentence', 'Options', 'Label'] ``` 3. **Check instructions**: Paper Section 3 specifies "Which of the following is the most analogous story?" → use exactly 4. **Generate scenario**: Paper instruction + dataset fields, MC pattern 5. **Verify**: 340 examples, Labels are A/B/C/D ### Example B: Standard format (RiddleSense) 1. **Qualify**: Creativity benchmark (riddles + commonsense)? Yes. Dataset? Yes. Eval? Yes. 2. **Examine dataset**: Fields are `question`, `choices`, `answerKey` 3. **Check instructions**: No specific wording in paper → use standard MC format, note in header 4. **Generate scenario**: Standard "Question: X\n\nA. ..." format 5. **Verify**: Test split has no labels → switch to validation split --- Source attribution: David-Li0406 Original project: https://github.com/David-Li0406/meta-skill-evloving/blob/ca3a335628981df10c36e00cb9850df2c247ab9a/skill-flow/data/skills-refined-36k/skillsmp/benchmark-onboarder/SKILL.md License: Apache-2.0 License information: https://www.apache.org/licenses/LICENSE-2.0 Original upstream instructions; downloaded from Mega AI Hub. ## Upstream license notice Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 1. Definitions. "License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document. "Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License. "Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity. "You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License. "Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files. "Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types. "Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below). "Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof. "Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution." "Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Work. 2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form. 3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed. 4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions: (a) You must give any other recipients of the Work or Derivative Works a copy of this License; and (b) You must cause any modified files to carry prominent notices stating that You changed the files; and (c) You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and (d) If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License. You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License. 5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions. 6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file. 7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License. 8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages. 9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability. END OF TERMS AND CONDITIONS APPENDIX: How to apply the Apache License to your work. To apply the Apache License to your work, attach the following boilerplate notice, with the fields enclosed by brackets "[]" replaced with your own identifying information. (Don't include the brackets!) The text should be enclosed in the appropriate comment syntax for the file format. We also recommend that a file or class name and description of purpose be included on the same "printed page" as the copyright notice for easier identification within third-party archives. Copyright [yyyy] [name of copyright owner] Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. # Mega AI Hub resource metadata Creator / publisher: David-Li0406 Original project: https://github.com/David-Li0406/meta-skill-evloving/blob/ca3a335628981df10c36e00cb9850df2c247ab9a/skill-flow/data/skills-refined-36k/skillsmp/benchmark-onboarder/SKILL.md Attribution basis: Source repository record Fields of work: Data & Analytics, Development & Engineering, Product Management Roles: Data Scientist, ML / AI Engineer, Product Manager Specialties: API Tools: Helm Country applicability: Country not specified Classification basis: Source catalog labels Instruction SHA-256: 2dc96ffdf2f4366bd8e875dc608e322dc146dc8bc3af9f50b35d70ca1a88f40f