Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
pandas
License:
Saibo-creator commited on
Commit
5ba58f8
·
verified ·
1 Parent(s): 7d5c73d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +33 -0
README.md CHANGED
@@ -300,6 +300,39 @@ task_categories:
300
 
301
  JSONSchemaBench is a benchmark of **real-world JSON schemas** designed to evaluate **structured output generation** for Large Language Models (LLMs). It contains approximately **10,000 JSON schemas**, capturing diverse constraints and complexities.
302
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
303
  ## 📌 Dataset Overview
304
  - **Purpose:** Evaluate the **efficiency** and **coverage** of structured output generation.
305
  - **Sources:** GitHub, Kubernetes, API specifications, curated collections.
 
300
 
301
  JSONSchemaBench is a benchmark of **real-world JSON schemas** designed to evaluate **structured output generation** for Large Language Models (LLMs). It contains approximately **10,000 JSON schemas**, capturing diverse constraints and complexities.
302
 
303
+ ## 📢 Important Update (March 10th, 2025)
304
+
305
+ We have restructured the dataset to include train/val/test splits. If you downloaded the dataset before this date, you might encounter errors like `KeyError: 'Github_easy'`.
306
+
307
+ To fix this issue, please follow one of the options below:
308
+
309
+ 1. Update How Subsets Are Accessed:
310
+ If you previously used:
311
+
312
+ ```python
313
+ from datasets import load_dataset, concatenate_datasets, DatasetDict, Dataset
314
+
315
+ subset: DatasetDict = load_dataset("epfl-dlab/JSONSchemaBench")
316
+ subset["Github_easy"]
317
+ ```
318
+ You can update it to:
319
+
320
+ ```python
321
+ from datasets import load_dataset, concatenate_datasets, DatasetDict, Dataset
322
+
323
+ subset: DatasetDict = load_dataset("epfl-dlab/JSONSchemaBench", name="Github_easy")
324
+ subset: Dataset = concatenate_datasets([subset["train"], subset["val"], subset["test"]])
325
+ ```
326
+
327
+ 2. Load the Dataset in the Old Structure:
328
+ If you need the previous structure, you can use a specific revision:
329
+
330
+ ```python
331
+ dataset = load_dataset("epfl-dlab/JSONSchemaBench", revision="e2ee5fdba65657c60d3a24b321172eb7141f8d73")
332
+ ```
333
+
334
+ We apologize for the inconvenience and appreciate your understanding! 😊
335
+
336
  ## 📌 Dataset Overview
337
  - **Purpose:** Evaluate the **efficiency** and **coverage** of structured output generation.
338
  - **Sources:** GitHub, Kubernetes, API specifications, curated collections.