example(dataset): ds z bench

example/datasets/LLM/z_bench/README.md

@@ -0,0 +1,32 @@
+---
+title: The `z-bench` Dataset
+---
+
+## The z-bench Dataset Description
+
+- [Homepage](https://github.com/zhenbench/z-bench)
+
+## The `z-bench` dataset Structure
+
+### Data Fields
+
+- `data` of type dict:
+    - `prompt`: `str`
+    - `task_type`: `str`
+    - `ref_answer`: `str`
+    - `gpt3d5`: `str`
+    - `gpt4`: `str`
+
+## Build `z-bench` sample Dataset locally
+
+```shell
+python3 dataset.py common
+```
+
+## Example
+
+Output the `10`th record of the `z-bench` dataset.
+
+```shell
+python3 example.py common
+```

example/datasets/LLM/z_bench/dataset.py

@@ -0,0 +1,44 @@
+import csv
+import sys
+from io import StringIO
+
+import requests
+
+from starwhale import Link, Image, dataset, MIMEType  # noqa: F401
+from starwhale.utils.retry import http_retry
+
+PATH_ROOT = "https://raw.githubusercontent.com/zhenbench/z-bench/main"
+
+
+@http_retry
+def request_link_text(index_link):
+    return requests.get(index_link, timeout=10).text
+
+
+def build_ds(ds_name: str) -> None:
+    ds = dataset(f"z_bench_{ds_name}", create="auto")
+    csv_reader = csv.reader(
+        StringIO(request_link_text(f"{PATH_ROOT}/{ds_name}.samples.csv")), delimiter=","
+    )
+    line_count = 0
+    for row in csv_reader:
+        if line_count == 0:
+            print(f'Column names are {", ".join(row)}')
+            line_count += 1
+        else:
+            ds.append(
+                {
+                    "prompt": row[0],
+                    "task_type": row[1],
+                    "ref_answer": row[2],
+                    "gpt3d5": row[3],
+                    "gpt4": row[4],
+                }
+            )
+
+    ds.commit()
+    ds.close()
+
+
+if __name__ == "__main__":
+    build_ds(sys.argv[1])

example/datasets/LLM/z_bench/example.py

@@ -0,0 +1,8 @@
+import sys
+
+from starwhale import dataset
+
+ds_name = sys.argv[1]
+ds = dataset(f"z_bench_{ds_name}/version/latest")
+row = ds[10]
+print(row.data)