Datasets:

ExpertCrowd
/

ec-swe

Tasks:

Text Generation

Modalities:

Formats:

Sub-tasks:

language-modeling

Languages:

Size:

Tags:

software-engineering

Libraries:

License:

Dataset card Data Studio Files Files and versions

servicesToptal commited on 21 days ago

Commit

5bd8791

·

verified ·

1 Parent(s): 66b737d

Update README.md

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -140,7 +140,7 @@ check (see [Curation Rationale](#curation-rationale)).
 - Evaluating coding agents on realistic repository-level tasks: give the agent
   `problem_statement` and a checkout of `repo` at `base_commit` with
   `setup_patch` applied, then score the resulting diff with the task's grader.
-- Studying difficulty: every task carries the per-run rewards of 128 graded
   agent runs, so results can be compared against a known baseline.
 - Building or validating graders and task-quality tooling, using the reference
   fix, the hidden tests and the weighted criteria together.

 - Evaluating coding agents on realistic repository-level tasks: give the agent
   `problem_statement` and a checkout of `repo` at `base_commit` with
   `setup_patch` applied, then score the resulting diff with the task's grader.
+- Studying difficulty: every task carries the per-run rewards of graded
   agent runs, so results can be compared against a known baseline.
 - Building or validating graders and task-quality tooling, using the reference
   fix, the hidden tests and the weighted criteria together.