[eval] Add the NUPA numeracy benchmark - #76
Draft
sklisa wants to merge 1 commit into
Draft
Conversation
Add a native NUPA benchmark with official-style numeric scoring, grouped metrics, and a streaming converter for publishing the nested source dataset as row-oriented Hugging Face data. Keep the production dataset repository as TODO_ORG/nupa-text-eval pending reviewer guidance and organization upload permission. Allow tokenizer-free OpenAI chat adapters to skip local context preflight so the benchmark can run through the supported API path. Part of #7297
penfever
requested changes
Aug 29, 2026
penfever
left a comment
penfever
left a comment
Collaborator
There was a problem hiding this comment.
🤖 Requesting changes: NUPA defaults to TODO_ORG/nupa-text-eval, so the documented non-debug command fails. Publish and pin the production flattened dataset, then replace the placeholder before merging. The required code CI also has not run.
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add NUPA / Number Cookbook as a native, non-agentic Evalchemy benchmark for
basic numeracy failure modes. The benchmark loads row-oriented examples,
requests numeric completions, applies a clean-room implementation of the
official text evaluator's observable parsing and scoring behavior, and reports
overall metrics with task-family and digit-length-bucket breakdowns.
Add a one-time streaming converter for the original nested
HaotongYang/NUPA_textdataset. The converter publishes flattened records and aprovenance-bearing dataset card without holding the complete source in memory.
The production repository remains
TODO_ORG/nupa-text-eval: this draft requestsreview of the design and permission to publish under a project-controlled
Hugging Face organization before replacing the placeholder.
Allow OpenAI chat adapters without a local tokenizer to skip endpoint token
preflight. A 40-example staging run completed through
openai-chat-completionswith no infrastructure-error responses; the stagingsubset is not suitable for reporting benchmark performance.
Part of #7297